Browse the corpus

General Internal Medicine 1 High Value Care Recommendations The American College of Physicians, in collaboration with syndromes, alveolar consolidation, and pneumothorax, multiple other organizations, is engaged in a worldwide and it is safer and less expensive than radiography and CT. initiative to promote the practice of High Value Care (HVC). The treatment of systemic exertion intolerance disease The goals of the HVC initiative are to improve health care focuses on nonpharmacologic therapies; pharmacologic outcomes by providing care of proven benefit and reducing therapies have not been shown to be consistently effective costs by avoiding unnecessary and even harmful interven- (see Item 91). tions. The initiative comprises several programs that inte- Polysomnography is usually unnecessary in the evaluation grate the important concept of health care value (balancing of insomnia unless the clinical presentation suggests a clinical benefit with costs and harms) for a given interven- sleep disorder. tion into a broad range of educational materials to address Cognitive behavioral therapy for insomnia is preferred to the needs of trainees, practicing physicians, and patients. pharmacologic therapy as first-line treatment for insomnia. Routine use of systemic antibiotics is not recommended HVC content has been integrated into MKSAP 19 in sev- for the treatment of venous stasis ulcers. eral important ways. MKSAP 19 includes HVC-identified In patients with medically unexplained symptoms, diag- key points in the text, HVC-focused multiple-choice ques- nostic testing should be limited, because unrevealing tions, and, in MKSAP Digital, an HVC custom quiz. From studies provide negligible reassurance, pose iatrogenic the text and questions, we have generated the following risks, and result in additional patient anxiety. list of HVC recommendations that meet the definition In patients with medically unexplained symptoms, it may below of high value care and bring us closer to our goal be more beneficial and value-driven to assess and treat of improving patient outcomes while conserving finite any potential underlying psychological symptoms and to resources. educate patients on successful coping skills (see Item 10). High Value Care Recommendation: A recommendation to An ECG should be obtained in all patients with syncope; choose diagnostic and management strategies for patients brain imaging, cardiac enzymes, and metabolic profile in specific clinical situations that balance clinical benefit have low diagnostic yield and should be performed only with cost and harms with the goal of improving patient when there is a moderate to high pretest probability of a outcomes. specific underlying condition. Patients with vasovagal syncope may be discharged from Below are the High Value Care Recommendations for the the emergency department without costly testing or General Internal Medicine 1 section of MKSAP 19. procedures (see Item 70). e Physicians are not obligated to provide diagnostic or ther- Compared with usual care, specialty palliative medicine apeutic interventions for which there is no evidence of improves overall quality of life, physical symptom clinical benefit. burden, mood, and caregiver satisfaction with patient Chest radiography is not indicated in the evaluation of care in the setting of serious illness. acute cough in the absence of abnormal vital signs or No specific dressing for venous ulcer stands out as supe- lung examination findings, unless there are other rior in direct comparisons; dry gauze should be avoided concerning clinical features (e.g., altered mental status). because it disrupts the formation of granulation tissue e Antibiotics are not recommended in patients with acute when removed (see Item 27). bronchitis or upper respiratory tract infection without a Enteral or parenteral artificial nutritional support at the clear bacterial cause. end of life does not improve survival, is invasive, and may e The mainstay of treatment of acute cough due to bron- cause side effects. chitis is patient education; antitussive agents have limited In the setting of serious illness, docusate sodium for efficacy (see Item 55). constipation is no better than placebo (see Item 112). ¢ The diagnostic test of choice for benign paroxysmal In patients with persistently debilitating dyspnea despite positional vertigo is the Dix-Hallpike maneuver; routine maximal medical therapy, nonpharmacologic treatment imaging is not recommended. strategies, such as pursed lip breathing, and devices ¢ Point-of-care lung ultrasonography is an accurate that enhance airflow are the least invasive and easiest to means of rapidly detecting pleural effusion, interstitial administer to reduce severity of symptoms (see Item 80).

The American College of Physicians, in collaboration with syndromes, alveolar consolidation, and pneumothorax, multiple other organizations, is engaged in a worldwide and it is safer and less expensive than radiography and CT. initiative to promote the practice of High Value Care (HVC). The treatment of systemic exertion intolerance disease The goals of the HVC initiative are to improve health care focuses on nonpharmacologic therapies; pharmacologic outcomes by providing care of proven benefit and reducing therapies have not been shown to be consistently effective costs by avoiding unnecessary and even harmful interven- (see Item 91). tions. The initiative comprises several programs that inte- Polysomnography is usually unnecessary in the evaluation grate the important concept of health care value (balancing of insomnia unless the clinical presentation suggests a clinical benefit with costs and harms) for a given interven- sleep disorder. tion into a broad range of educational materials to address Cognitive behavioral therapy for insomnia is preferred to the needs of trainees, practicing physicians, and patients. pharmacologic therapy as first-line treatment for insomnia. Routine use of systemic antibiotics is not recommended HVC content has been integrated into MKSAP 19 in sev- for the treatment of venous stasis ulcers. eral important ways. MKSAP 19 includes HVC-identified In patients with medically unexplained symptoms, diag- key points in the text, HVC-focused multiple-choice ques- nostic testing should be limited, because unrevealing tions, and, in MKSAP Digital, an HVC custom quiz. From studies provide negligible reassurance, pose iatrogenic the text and questions, we have generated the following risks, and result in additional patient anxiety. list of HVC recommendations that meet the definition In patients with medically unexplained symptoms, it may below of high value care and bring us closer to our goal be more beneficial and value-driven to assess and treat of improving patient outcomes while conserving finite any potential underlying psychological symptoms and to resources. educate patients on successful coping skills (see Item 10). High Value Care Recommendation: A recommendation to An ECG should be obtained in all patients with syncope; choose diagnostic and management strategies for patients brain imaging, cardiac enzymes, and metabolic profile in specific clinical situations that balance clinical benefit have low diagnostic yield and should be performed only with cost and harms with the goal of improving patient when there is a moderate to high pretest probability of a outcomes. specific underlying condition. Patients with vasovagal syncope may be discharged from Below are the High Value Care Recommendations for the the emergency department without costly testing or General Internal Medicine 1 section of MKSAP 19. procedures (see Item 70). e Physicians are not obligated to provide diagnostic or ther- Compared with usual care, specialty palliative medicine apeutic interventions for which there is no evidence of improves overall quality of life, physical symptom clinical benefit. burden, mood, and caregiver satisfaction with patient Chest radiography is not indicated in the evaluation of care in the setting of serious illness. acute cough in the absence of abnormal vital signs or No specific dressing for venous ulcer stands out as supe- lung examination findings, unless there are other rior in direct comparisons; dry gauze should be avoided concerning clinical features (e.g., altered mental status). because it disrupts the formation of granulation tissue e Antibiotics are not recommended in patients with acute when removed (see Item 27). bronchitis or upper respiratory tract infection without a Enteral or parenteral artificial nutritional support at the clear bacterial cause. end of life does not improve survival, is invasive, and may e The mainstay of treatment of acute cough due to bron- cause side effects. chitis is patient education; antitussive agents have limited In the setting of serious illness, docusate sodium for efficacy (see Item 55). constipation is no better than placebo (see Item 112). ¢ The diagnostic test of choice for benign paroxysmal In patients with persistently debilitating dyspnea despite positional vertigo is the Dix-Hallpike maneuver; routine maximal medical therapy, nonpharmacologic treatment imaging is not recommended. strategies, such as pursed lip breathing, and devices ¢ Point-of-care lung ultrasonography is an accurate that enhance airflow are the least invasive and easiest to means of rapidly detecting pleural effusion, interstitial administer to reduce severity of symptoms (see Item 80). xi

¢ Compared with second-generation antidepressants, psy- the patella and classically occurs with activities that chostimulants (dextroamphetamine, methylphenidate, involve knee flexion; imaging is not necessary to establish and pemoline) should be considered in the treatment the diagnosis (see Item 50). of depression at the end of life because they take effect The diagnosis of iliotibial band syndrome can be made quickly (see Item 9). when repeated flexion and extension of the supine e Most patients with nonspecific low back pain do not patient’s knee with the thumb over the lateral femoral require imaging or other diagnostic testing. condyle reproduces the patient’s pain; imaging is not ¢ Nonpharmacologic therapies are considered a cornerstone necessary (see Item 28). of treatment for acute and chronic low back pain; opioid In patients treated with statin therapy, liver chemistry therapy should be avoided. tests and muscle enzyme studies should not be routinely Patient education can provide long-term reassurance obtained in the absence of symptoms of liver dysfunction for patients with acute or subacute low back pain and or myopathy. reduces low back pain-related primary care visits more In the treatment of hypertriglyceridemia, evidence of a than usual care (see Item 99). beneficial effect for omega-3 fatty acid supplements other Most patients with neck pain do not require imaging than icosapent ethyl is lacking (see Item 60). studies; plain radiography should be performed when For initial acute treatment of major depressive disorder, evaluation suggests fracture, infection, or cancer and may options include cognitive behavioral therapy or second- be considered for cervical sprain that is unresponsive to 6 generation antidepressants after consideration and dis- to 8 weeks of conservative therapy. cussion of side effects, cost, accessibility, and patient Most patients with rotator cuff disease do not require preferences; combination therapy is a reasonable option, imaging studies or surgery; treatment may include edu- especially in moderate to severe disease. cation, avoidance or modification of aggravating activi- Nonpharmacologic therapy, including lifestyle modifica- ties, physical therapy, and acetaminophen or NSAIDs. tion and behavioral therapy, is the preferred treatment for Nonsurgical management of olecranon bursitis is signifi- all types of urinary incontinence. cantly more effective than surgical management, leading Alternating air mattresses for pressure injury prevention to higher rates of clinical resolution (see Item 30). are not recommended, based on cost considerations and e In patients with a suspected scaphoid fracture but negative no evidence of clear advantage (see Item 98). radiograph, immediate MRI or CT is more cost-effective Evidence is insufficient to assess the safety and efficacy than splinting and repeat radiography when follow-up of hyperbaric oxygen therapy for pressure injuries (see care, missed fracture, and lost productivity are taken into Item 51). account (see Item 19). Actinic purpura is the result of blood vessel fragility and ¢ Most medial collateral ligament and posterior cruciate dermal atrophy from aging; no additional testing needs to ligament tears do not require surgery and can be managed be done, and no treatment is required. conservatively. Results on direct-to-consumer genomic testing should be e Patellofemoral pain syndrome is characterized by pain in confirmed with a clinical test before being used for medi- the knee that may be localized to the patella or behind cal decision making (see Item 73).

¢ Compared with second-generation antidepressants, psy- the patella and classically occurs with activities that chostimulants (dextroamphetamine, methylphenidate, involve knee flexion; imaging is not necessary to establish and pemoline) should be considered in the treatment the diagnosis (see Item 50). of depression at the end of life because they take effect The diagnosis of iliotibial band syndrome can be made quickly (see Item 9). when repeated flexion and extension of the supine e Most patients with nonspecific low back pain do not patient’s knee with the thumb over the lateral femoral require imaging or other diagnostic testing. condyle reproduces the patient’s pain; imaging is not ¢ Nonpharmacologic therapies are considered a cornerstone necessary (see Item 28). of treatment for acute and chronic low back pain; opioid In patients treated with statin therapy, liver chemistry therapy should be avoided. tests and muscle enzyme studies should not be routinely Patient education can provide long-term reassurance obtained in the absence of symptoms of liver dysfunction for patients with acute or subacute low back pain and or myopathy. reduces low back pain-related primary care visits more In the treatment of hypertriglyceridemia, evidence of a than usual care (see Item 99). beneficial effect for omega-3 fatty acid supplements other Most patients with neck pain do not require imaging than icosapent ethyl is lacking (see Item 60). studies; plain radiography should be performed when For initial acute treatment of major depressive disorder, evaluation suggests fracture, infection, or cancer and may options include cognitive behavioral therapy or second- be considered for cervical sprain that is unresponsive to 6 generation antidepressants after consideration and dis- to 8 weeks of conservative therapy. cussion of side effects, cost, accessibility, and patient Most patients with rotator cuff disease do not require preferences; combination therapy is a reasonable option, imaging studies or surgery; treatment may include edu- especially in moderate to severe disease. cation, avoidance or modification of aggravating activi- Nonpharmacologic therapy, including lifestyle modifica- ties, physical therapy, and acetaminophen or NSAIDs. tion and behavioral therapy, is the preferred treatment for Nonsurgical management of olecranon bursitis is signifi- all types of urinary incontinence. cantly more effective than surgical management, leading Alternating air mattresses for pressure injury prevention to higher rates of clinical resolution (see Item 30). are not recommended, based on cost considerations and e In patients with a suspected scaphoid fracture but negative no evidence of clear advantage (see Item 98). radiograph, immediate MRI or CT is more cost-effective Evidence is insufficient to assess the safety and efficacy than splinting and repeat radiography when follow-up of hyperbaric oxygen therapy for pressure injuries (see care, missed fracture, and lost productivity are taken into Item 51). account (see Item 19). Actinic purpura is the result of blood vessel fragility and ¢ Most medial collateral ligament and posterior cruciate dermal atrophy from aging; no additional testing needs to ligament tears do not require surgery and can be managed be done, and no treatment is required. conservatively. Results on direct-to-consumer genomic testing should be e Patellofemoral pain syndrome is characterized by pain in confirmed with a clinical test before being used for medi- the knee that may be localized to the patella or behind cal decision making (see Item 73). xii

General Internal Medicine 1 High Value Care implemented reimbursement models quality, efficient care. that reward high- in Internal Medicine When determining the value of an intervention, clini- cians should consider benefits before cost, because focusing The United States spends more on health care than all other primarily on cost might result in continuing low-cost inter- developed nations; however, it has higher rates of medical ventions that provide no benefit or avoiding expensive but care-related mortality and shorter life expectancy. In response highly beneficial interventions. For example, some costly or to this apparent paradox, health policy organizations advocate implementing a high-value approach to patient care. High risky interventions, such as screening colonoscopy, are consid- value care is individualized care that delivers proven benefits ered high value, whereas some inexpensive and low-risk inter- while minimizing risks and unnecessary costs; it eliminates ventions, such as daily laboratory testing in hospitalized wasteful health care habits, improves health outcomes, and patients, represent low value care. Examples of high value and avoids harms. Third-party payers, including Medicare, have low value care are provided in Table 1. TABLE 1. Examples of High Value and Low Value Care Intervention Cost Benefit High Value Care

wasteful health care habits, improves health outcomes, and patients, represent low value care. Examples of high value and avoids harms. Third-party payers, including Medicare, have low value care are provided in Table 1. TABLE 1. Examples of High Value and Low Value Care Intervention Cost Benefit High Value Care High-sensitivity D-dimer testing rather than Low High negative predictive value imaging studies to exclude venous thromboem- bolism in patients with low likelihood of disease Influenza vaccination Low High benefit for reducing disease burden and complications Dix-Hallpike maneuver as the diagnostic test of Low Highly accurate diagnostic test with few choice for benign paroxysmal positional vertigo or no harms Cognitive behavioral therapy for insomnia Intermediate Effective and may avoid risks associated with pharmacologic therapy Hydrocolloid or foam dressings rather than Intermediate Improved rate of ulcer healing and standard gauze dressings in the treatment of reduction of pain pressure injuries Low Value Care Antibiotic therapy in patients with upper Low No benefit for reducing duration or respiratory tract infection severity of illness Stress ulcer prophylaxis in patients at low or Low No benefit for reducing clinical events moderate risk

Antibiotic therapy in patients with upper Low No benefit for reducing duration or respiratory tract infection severity of illness Stress ulcer prophylaxis in patients at low or Low No benefit for reducing clinical events moderate risk Obtaining an ECG in adults at lowrisk for Low initial cost but increased downstream No effect on clinical outcomes cardiovascular disease costs with unnecessary consultations, echocardiography, and stress testing Pap testing in low-risk women (aged Low initial cost but increased downstream No effect on clinical outcomes 13-20 y or >65 y) costs with unnecessary consultations and repeat or additional testing Carotid ultrasonography in patients with syncope Intermediate Low diagnostic value | Chest radiography in adults without Intermediate initial cost but increased No effect on clinical outcomes cardiovascular or pulmonary disease downstream costs with unnecessary consultations and CT Imaging studies in patients with low back pain Intermediate Low diagnostic value in the absence of "red flag” findings (fever, involuntary weight loss, incontinence) Routine preoperative testing before low-risk Intermediate No effect on clinical outcomes surgery

Clinical Decision Making and Interpreting the Medical Literature High value care places patients and their outcomes, val- approach, such as hypothetico-deductive reasoning, in ues, and concerns at the center of every major clinical decision which each diagnostic possibility, including the concordant and involves patients in decision making. Perceived patient and discordant aspects of the clinical presentation, is spe- demand is one important driver of inappropriate testing and cifically analyzed. System 1 processes have the advantage of treatment, and including patients in the decision-making pro- efficiency but are also prone to error. Many clinical decisions cess provides opportunities to educate them on balancing involve the use of both systems, and it is difficult to ascertain benefits with costs and to incorporate their values and prefer- which system a clinician is using. Experienced clinicians, ences into the care plan. however, tend to utilize system 1 more than inexperienced Evidence-based decision support tools that consider clinicians and often use system 2 to check their system 1 value may assist physicians in providing high value care. In processes.

High value care places patients and their outcomes, val- approach, such as hypothetico-deductive reasoning, in ues, and concerns at the center of every major clinical decision which each diagnostic possibility, including the concordant and involves patients in decision making. Perceived patient and discordant aspects of the clinical presentation, is spe- demand is one important driver of inappropriate testing and cifically analyzed. System 1 processes have the advantage of treatment, and including patients in the decision-making pro- efficiency but are also prone to error. Many clinical decisions cess provides opportunities to educate them on balancing involve the use of both systems, and it is difficult to ascertain benefits with costs and to incorporate their values and prefer- which system a clinician is using. Experienced clinicians, ences into the care plan. however, tend to utilize system 1 more than inexperienced Evidence-based decision support tools that consider clinicians and often use system 2 to check their system 1 value may assist physicians in providing high value care. In processes. addition, the American College of Physicians offers resources to help clinicians determine the value of common tests and interventions (www.acponline.org/clinical-information/ Diagnostic Reasoning and high-value-care). The Choosing Wisely campaign, an initia- Diagnostic Error tive of the American Board of Internal Medicine Foundation Diagnostic reasoning is a multifaceted process that involves with medical specialty societies, has published specialty- systematically obtaining historical information, observing specific lists of commonly used tests or procedures whose physical findings, evaluating diagnostic test results, and necessity should be questioned and discussed by clinicians applying the information obtained to categorize the clinical and patients (www.choosingwisely.org). Patient-friendly syndrome accurately. The process comprises multiple inter- materials regarding these recommendations are available to related steps (Figure 1). Using the data acquired, the clinician consumers through a network of Choosing Wisely partners. summarizes the important clinical features as a problem rep- Care coordination programs that focus on patients at risk for resentation. This step guides hypothesis generation through high-cost care also may facilitate the provision of high value matching of the problem representation to illness scripts, or care. mental representations of diseases and conditions that a clini- cian builds over time through experience. Activated illness scripts drive the acquisition of more data that increase or ¢ High value care is individualized care that delivers decrease the likelihood of specific diseases. Interpreting the proven benefits while minimizing risks and unnecessary clinical information to arrive at the correct diagnosis requires costs. an understanding of how specific features or test results influ- ence the probability of a disease or condition. Therapeutic reasoning operates in a similar manner; appropriate treat-

addition, the American College of Physicians offers resources to help clinicians determine the value of common tests and interventions (www.acponline.org/clinical-information/ Diagnostic Reasoning and high-value-care). The Choosing Wisely campaign, an initia- Diagnostic Error tive of the American Board of Internal Medicine Foundation Diagnostic reasoning is a multifaceted process that involves with medical specialty societies, has published specialty- systematically obtaining historical information, observing specific lists of commonly used tests or procedures whose physical findings, evaluating diagnostic test results, and necessity should be questioned and discussed by clinicians applying the information obtained to categorize the clinical and patients (www.choosingwisely.org). Patient-friendly syndrome accurately. The process comprises multiple inter- materials regarding these recommendations are available to related steps (Figure 1). Using the data acquired, the clinician consumers through a network of Choosing Wisely partners. summarizes the important clinical features as a problem rep- Care coordination programs that focus on patients at risk for resentation. This step guides hypothesis generation through high-cost care also may facilitate the provision of high value matching of the problem representation to illness scripts, or care. mental representations of diseases and conditions that a clini- cian builds over time through experience. Activated illness scripts drive the acquisition of more data that increase or ¢ High value care is individualized care that delivers decrease the likelihood of specific diseases. Interpreting the proven benefits while minimizing risks and unnecessary clinical information to arrive at the correct diagnosis requires costs. an understanding of how specific features or test results influ- ence the probability of a disease or condition. Therapeutic reasoning operates in a similar manner; appropriate treat- Clinical Decision Making ment is selected by applying evidence from the medical litera- ture to the unique characteristics and preferences of the and Interpreting the individual patient.

Clinical Decision Making ment is selected by applying evidence from the medical litera- ture to the unique characteristics and preferences of the and Interpreting the individual patient. Medical Literature Knowledge Clinical Decision Making Patient's story TM * Clinical decision making, or the application of reasoning to a Vv clinical situation, is an essential skill in clinical practice, medi- Data acquisition <—————__ ! cal education, and research. Despite its ubiquity, experts often disagree on how to define this process. Some definitions limit Accurate “problem representation” decision making to the individual cognitive process of the cli- ! Context nician, whereas other definitions include social, environmen- tal, and technical influences in the decision-making process. Generation of hypothesis Much of the research in this area has focused on diagnostic reasoning, because diagnostic error is a leading contributor to preventable harm. However, in practice, clinical decision mak- Search for and selection <——— ing can be applied to diagnostic, therapeutic, screening, and of illness script

tal, and technical influences in the decision-making process. Generation of hypothesis Much of the research in this area has focused on diagnostic reasoning, because diagnostic error is a leading contributor to preventable harm. However, in practice, clinical decision mak- Search for and selection <——— ing can be applied to diagnostic, therapeutic, screening, and of illness script advance care planning decisions. Experience a Y Clinicians primarily use two parallel and intertwined Diagnosis systems when making decisions, known as dual-process theory. System 1 entails an intuitive, pattern-based approach, FIGURE 1. Key elements of the clinical diagnostic reasoning process. such as the instant recognition of a vesicular rash as herpes Reproduced with permission from Bowen JL. Educational strategies to promote clinical diagnostic reasoning. N EnglJ Med. 2006;355:2217-25. [PMID: 17124019] doi:10.1056/NEJMra054782 ©2006, Massachusetts zoster. In contrast, system 2 uses an analytical, rule-based Medical Society. 2

Clinical Decision Making and Interpreting the Medical Literature Experience is the most important factor in the devel- pressures and administrative burden, and their effect on clini- opment and improvement of diagnostic reasoning abili- cian time with patients, may also contribute to diagnostic ties. The effectiveness of other proposed means of errors. Chaos in a medical practice is associated with an improvement is unclear; however, guided reflection and increased risk for error, missed opportunities to provide cognitive forcing strategies have shown promise in improv- preventive services, burnout, and physician turnover. It is ing the accuracy of clinical decision making in early- essential for institutions to implement strategies to address career physicians. Cognitive forcing strategies are specific systematic contributors to diagnostic error through protocols, techniques involving self-monitoring of decision making policies, resource allocation, and quality initiatives. to prevent an inappropriate pattern recognition diagnosis that leads to diagnostic error. Other interventions that may Interpreting and Applying the be beneficial include checklists and educational sessions. Medical Literature There are fewer data on effective interventions for mid- or Effective clinical decision making relies on an ability to ana- late-career physicians. The Clinical Reasoning Toolkit lyze the available medical literature and translate the evidence from the Society to Improve Diagnosis in Medicine offers a variety of resources for improving diagnostic decision mak- base for use in individual clinical scenarios. To accomplish this, it is important to understand which study designs are best ing (www.improvediagnosis.org/clinicalreasoning/). suited to answer the specific clinical question (e.g., etiology, diagnosis, treatment, prevention, or prognosis), how to inter- Diagnostic Error pret statistical tests for significance, and how to determine Diagnostic error occurs when a patient is not provided with which information is useful in patient-centered clinical deci- a timely and correct diagnosis or the diagnosis is not commu- sion making.

Experience is the most important factor in the devel- pressures and administrative burden, and their effect on clini- opment and improvement of diagnostic reasoning abili- cian time with patients, may also contribute to diagnostic ties. The effectiveness of other proposed means of errors. Chaos in a medical practice is associated with an improvement is unclear; however, guided reflection and increased risk for error, missed opportunities to provide cognitive forcing strategies have shown promise in improv- preventive services, burnout, and physician turnover. It is ing the accuracy of clinical decision making in early- essential for institutions to implement strategies to address career physicians. Cognitive forcing strategies are specific systematic contributors to diagnostic error through protocols, techniques involving self-monitoring of decision making policies, resource allocation, and quality initiatives. to prevent an inappropriate pattern recognition diagnosis that leads to diagnostic error. Other interventions that may Interpreting and Applying the be beneficial include checklists and educational sessions. Medical Literature There are fewer data on effective interventions for mid- or Effective clinical decision making relies on an ability to ana- late-career physicians. The Clinical Reasoning Toolkit lyze the available medical literature and translate the evidence from the Society to Improve Diagnosis in Medicine offers a variety of resources for improving diagnostic decision mak- base for use in individual clinical scenarios. To accomplish this, it is important to understand which study designs are best ing (www.improvediagnosis.org/clinicalreasoning/). suited to answer the specific clinical question (e.g., etiology, diagnosis, treatment, prevention, or prognosis), how to inter- Diagnostic Error pret statistical tests for significance, and how to determine Diagnostic error occurs when a patient is not provided with which information is useful in patient-centered clinical deci- a timely and correct diagnosis or the diagnosis is not commu- sion making. nicated to the patient. More specifically, a diagnostic error is Although applying the evidence may seem like a daunting

Experience is the most important factor in the devel- pressures and administrative burden, and their effect on clini- opment and improvement of diagnostic reasoning abili- cian time with patients, may also contribute to diagnostic ties. The effectiveness of other proposed means of errors. Chaos in a medical practice is associated with an improvement is unclear; however, guided reflection and increased risk for error, missed opportunities to provide cognitive forcing strategies have shown promise in improv- preventive services, burnout, and physician turnover. It is ing the accuracy of clinical decision making in early- essential for institutions to implement strategies to address career physicians. Cognitive forcing strategies are specific systematic contributors to diagnostic error through protocols, techniques involving self-monitoring of decision making policies, resource allocation, and quality initiatives. to prevent an inappropriate pattern recognition diagnosis that leads to diagnostic error. Other interventions that may Interpreting and Applying the be beneficial include checklists and educational sessions. Medical Literature There are fewer data on effective interventions for mid- or Effective clinical decision making relies on an ability to ana- late-career physicians. The Clinical Reasoning Toolkit lyze the available medical literature and translate the evidence from the Society to Improve Diagnosis in Medicine offers a variety of resources for improving diagnostic decision mak- base for use in individual clinical scenarios. To accomplish this, it is important to understand which study designs are best ing (www.improvediagnosis.org/clinicalreasoning/). suited to answer the specific clinical question (e.g., etiology, diagnosis, treatment, prevention, or prognosis), how to inter- Diagnostic Error pret statistical tests for significance, and how to determine Diagnostic error occurs when a patient is not provided with which information is useful in patient-centered clinical deci- a timely and correct diagnosis or the diagnosis is not commu- sion making. nicated to the patient. More specifically, a diagnostic error is Although applying the evidence may seem like a daunting present when there was a missed opportunity to make the task, clinical probability tools, likelihood ratios, and numbers correct diagnosis in a timely manner. The rate of diagnostic needed to treat and harm provide a useful means of doing so. error is approximately 10%, and most patients will be subject As an example, the Pulmonary Embolism Rule-out Criteria to a diagnostic error at some point in their lifetime. Diagnostic identify a specific clinical population with several clinical

present when there was a missed opportunity to make the task, clinical probability tools, likelihood ratios, and numbers correct diagnosis in a timely manner. The rate of diagnostic needed to treat and harm provide a useful means of doing so. error is approximately 10%, and most patients will be subject As an example, the Pulmonary Embolism Rule-out Criteria to a diagnostic error at some point in their lifetime. Diagnostic identify a specific clinical population with several clinical errors are preventable in half or more of cases, making them findings whose risk for pulmonary embolism is sufficiently the leading cause of malpractice claims against internists. low enough that diagnostic testing is not warranted.

present when there was a missed opportunity to make the task, clinical probability tools, likelihood ratios, and numbers correct diagnosis in a timely manner. The rate of diagnostic needed to treat and harm provide a useful means of doing so. error is approximately 10%, and most patients will be subject As an example, the Pulmonary Embolism Rule-out Criteria to a diagnostic error at some point in their lifetime. Diagnostic identify a specific clinical population with several clinical errors are preventable in half or more of cases, making them findings whose risk for pulmonary embolism is sufficiently the leading cause of malpractice claims against internists. low enough that diagnostic testing is not warranted. Diagnostic errors are usually multifactorial in etiology, with flaws in the individual clinician’s cognitive processes and Study Designs systems-level issues contributing to most errors. Examples of Study designs can be divided into two broad categories: cognitive errors include premature closure, diagnostic experimental studies and observational studies. Study design momentum, confirmation bias, and faulty knowledge or selection is guided by the population being studied and the knowledge application. Premature closure involves accepting suitability and feasibility of the design to address the clinical a diagnosis and discontinuing the diagnostic process before question. Experimental studies, especially those with random the data necessary to establish the diagnosis have been allocation of treatment, provide the strongest evidence for obtained. Diagnostic momentum is a similar phenomenon in determining the etiology of an outcome, effectiveness of pre- which a diagnosis is suggested early in the diagnostic process, ventive measures, and effectiveness of treatments or thera- and the diagnostic evaluation continues to move toward that pies. For studies of diagnosis or prognosis, the observational diagnosis even if the data do not support it. Confirmation bias, prospective cohort study often offers the strongest available the predisposition to seek evidence to confirm a suspected evidence. diagnosis without looking for evidence that disproves it, is a Levels of evidence describe the strength and quality of common feature of premature closure and diagnostic momen- study results and can aid in clinical decision making. tum. Faulty application of knowledge occurs when the clini- Systematic reviews, with or without meta-analysis, provide cian does not possess the underlying knowledge necessary to the highest level of evidence, followed by large, multicenter make the diagnosis or does not apply the knowledge properly. randomized, blinded, placebo-controlled trials. Large, meticu- Techniques that clinicians can use to reduce the risk for diag- lously controlled studies generally provide a higher level of nostic error are provided in Table 2. evidence than smaller studies, and experimental studies pro- Systems factors that contribute to diagnostic error include vide a higher level of evidence than observational studies. communication policies and practices, resource availability, Reports of expert opinion provide the lowest acceptable level and practice structure. Poor communication of crucial labora- of evidence. The U.S. Preventive Services Task Force and other tory and imaging results (e.g., the incidental finding of a pul- organizations have developed grading systems to rank the rela- monary nodule) may result in delayed diagnosis and patient tive rigor of clinical studies (Table 3). harm. Similarly, lack of test availability (e.g., advanced imag- Recommendations in clinical practice guidelines gener- ing in a rural location) may delay diagnosis. Productivity ally reflect the certainty of evidence as well as the class (or

Diagnostic errors are usually multifactorial in etiology, with flaws in the individual clinician’s cognitive processes and Study Designs systems-level issues contributing to most errors. Examples of Study designs can be divided into two broad categories: cognitive errors include premature closure, diagnostic experimental studies and observational studies. Study design momentum, confirmation bias, and faulty knowledge or selection is guided by the population being studied and the knowledge application. Premature closure involves accepting suitability and feasibility of the design to address the clinical a diagnosis and discontinuing the diagnostic process before question. Experimental studies, especially those with random the data necessary to establish the diagnosis have been allocation of treatment, provide the strongest evidence for obtained. Diagnostic momentum is a similar phenomenon in determining the etiology of an outcome, effectiveness of pre- which a diagnosis is suggested early in the diagnostic process, ventive measures, and effectiveness of treatments or thera- and the diagnostic evaluation continues to move toward that pies. For studies of diagnosis or prognosis, the observational diagnosis even if the data do not support it. Confirmation bias, prospective cohort study often offers the strongest available the predisposition to seek evidence to confirm a suspected evidence. diagnosis without looking for evidence that disproves it, is a Levels of evidence describe the strength and quality of common feature of premature closure and diagnostic momen- study results and can aid in clinical decision making. tum. Faulty application of knowledge occurs when the clini- Systematic reviews, with or without meta-analysis, provide cian does not possess the underlying knowledge necessary to the highest level of evidence, followed by large, multicenter make the diagnosis or does not apply the knowledge properly. randomized, blinded, placebo-controlled trials. Large, meticu- Techniques that clinicians can use to reduce the risk for diag- lously controlled studies generally provide a higher level of nostic error are provided in Table 2. evidence than smaller studies, and experimental studies pro- Systems factors that contribute to diagnostic error include vide a higher level of evidence than observational studies. communication policies and practices, resource availability, Reports of expert opinion provide the lowest acceptable level and practice structure. Poor communication of crucial labora- of evidence. The U.S. Preventive Services Task Force and other tory and imaging results (e.g., the incidental finding of a pul- organizations have developed grading systems to rank the rela- monary nodule) may result in delayed diagnosis and patient tive rigor of clinical studies (Table 3). harm. Similarly, lack of test availability (e.g., advanced imag- Recommendations in clinical practice guidelines gener- ing in a rural location) may delay diagnosis. Productivity ally reflect the certainty of evidence as well as the class (or 3

Clinical Decision Making and Interpreting the Medical Literature TABLE 2. Twelve Tips for Avoiding Diagnostic Errors | Technique Comments te Understand heuristics? Availability heuristic: Diagnosing based on what is most easily available in the physician's memory (e.g., because of a patient recently seen) rather than what is most probable Anchoring heuristic: Settling on a diagnosis early in the diagnostic process despite data that refute the diagnosis or support another diagnosis (premature closure) Representativeness heuristic: Application of pattern recognition (a patient's presentation fits a “typical” case; therefore, it must be that case) | . Use “diagnostic timeouts” Taking time to periodically review a case on the basis of data but without assuming that the | diagnosis is that which was previously reached . Practice “worst-case scenario Consider the most life-threatening diagnoses first: medicine” ¢ Lessens chance of missing these diagnoses © Does not mandate testing for them . Use a systematic approach to For example, anatomic approach to abdominal pain beginning from exterior to interior common problems . Ask why For example, when a patient presents with diabetic ketoacidosis or a COPD exacerbation, ask what prompted this acute exacerbation of a chronic condition 6. Use the clinical examination Decreases reliance on a single test and decreases the chance of premature closure 7. Use Bayes theorem Use pre- and posttest probabilities © Helps avert premature closure based on a single test result

. Ask why For example, when a patient presents with diabetic ketoacidosis or a COPD exacerbation, ask what prompted this acute exacerbation of a chronic condition 6. Use the clinical examination Decreases reliance on a single test and decreases the chance of premature closure 7. Use Bayes theorem Use pre- and posttest probabilities © Helps avert premature closure based on a single test result . Acknowledge the effect of the How does the patient make the physician feel? patient on clinical thinking ¢ Physicians may avoid making unfavorable diagnoses in patients with whom they identify ¢ Physicians may discount important data in patients with whom they have difficult encounters . Look for clinical findings that do Encourages a comprehensive approach and incorporates healthy skepticism not fit the diagnosis 10. Consider “zebras” Resist the temptation to lock onto common diagnoses at the risk of missing the uncommon thie Slow down and reflect Difficult to do in most health care systems, which stress the economy of “getting it right the first time” 12 Admit mistakes Awareness of one’s own fallibility may lead to fewer diagnostic errors later *Heuristics are shortcuts in reasoning used in discovery, learning, or problem solving. Information from Trowbridge RL. Twelve tips for teaching avoidance of diagnostic errors. Med Teach. 2008;30:496-500. [PMID: 18576188] doi:10.1080/01421590801965137 |

12 Admit mistakes Awareness of one’s own fallibility may lead to fewer diagnostic errors later *Heuristics are shortcuts in reasoning used in discovery, learning, or problem solving. Information from Trowbridge RL. Twelve tips for teaching avoidance of diagnostic errors. Med Teach. 2008;30:496-500. [PMID: 18576188] doi:10.1080/01421590801965137 | strength) of recommendation, which considers the balance of benefits versus risks for the intervention. A commonly used TABLE 3. U.S. Preventive Services Task Force Hierarchy of grading system for class of recommendation is presented in Research Design Figure 2. | Level Description Properly powered and conducted RCT; well- Experimental Studies conducted systematic review or meta-analysis of homogeneous RCTs Experimental studies systematically expose some or all of the individuals in the study to a potential causative or protective i +1 Well-designed controlled trial without randomization factor (the exposure, often a treatment). Exposed and unex- WI-2 Well-designed cohort or case-control analytic study posed participants are compared with respect to developing II-3 Multiple time series with or without the intervention; | the outcome of interest. Examples of experimental study results from uncontrolled studies that yield results of large magnitude | designs are provided in Table 4. The process for allocating the exposure or treatment to I Opinions of respected authorities, based on clinical | experience; descriptive studies or case reports; the study participants affects the strength and validity of the

strength) of recommendation, which considers the balance of benefits versus risks for the intervention. A commonly used TABLE 3. U.S. Preventive Services Task Force Hierarchy of grading system for class of recommendation is presented in Research Design Figure 2. | Level Description Properly powered and conducted RCT; well- Experimental Studies conducted systematic review or meta-analysis of homogeneous RCTs Experimental studies systematically expose some or all of the individuals in the study to a potential causative or protective i +1 Well-designed controlled trial without randomization factor (the exposure, often a treatment). Exposed and unex- WI-2 Well-designed cohort or case-control analytic study posed participants are compared with respect to developing II-3 Multiple time series with or without the intervention; | the outcome of interest. Examples of experimental study results from uncontrolled studies that yield results of large magnitude | designs are provided in Table 4. The process for allocating the exposure or treatment to I Opinions of respected authorities, based on clinical | experience; descriptive studies or case reports; the study participants affects the strength and validity of the reports of expert committees | study. The strongest experimental design is the randomized RCT = randomized controlled trial. | controlled trial. By randomly allocating the treatment, con-

strength) of recommendation, which considers the balance of benefits versus risks for the intervention. A commonly used TABLE 3. U.S. Preventive Services Task Force Hierarchy of grading system for class of recommendation is presented in Research Design Figure 2. | Level Description Properly powered and conducted RCT; well- Experimental Studies conducted systematic review or meta-analysis of homogeneous RCTs Experimental studies systematically expose some or all of the individuals in the study to a potential causative or protective i +1 Well-designed controlled trial without randomization factor (the exposure, often a treatment). Exposed and unex- WI-2 Well-designed cohort or case-control analytic study posed participants are compared with respect to developing II-3 Multiple time series with or without the intervention; | the outcome of interest. Examples of experimental study results from uncontrolled studies that yield results of large magnitude | designs are provided in Table 4. The process for allocating the exposure or treatment to I Opinions of respected authorities, based on clinical | experience; descriptive studies or case reports; the study participants affects the strength and validity of the reports of expert committees | study. The strongest experimental design is the randomized RCT = randomized controlled trial. | controlled trial. By randomly allocating the treatment, con- Reproduced from U.S. Preventive Services Task Force. Procedure Manual. | founders (both known and unknown) are randomly distrib- | www.uspreventiveservicestaskforce.org/Page/Name/procedure-manual. | uted in the treatment and control groups. The least rigorous December 2015. Accessed March 2, 2020. | experimental design is the quasi-experimental design, in

Clinical Decision Making and Interpreting the Medical Literature Grading Certainty of Evidence High Confident that the true effect is close to the estimated effect. Moderate Moderately confident in the effect estimate: The true effect is likely close to the estimated effect, but there is a sizable possibility that it is substantially different. Low Confidence in the effect estimate is limited: The true effect may be substantially different from the estimated effect. Grading Recommendations Strength Certainty of Evidence | Balance of Benefits and | Applicable Patient Policy Implication Harms Population Strong High or moderate Confidence that benefits | Applies to most patients | Only strong recommendations could be considered clearly outweigh risks in most circumstances. for use as performance measures. Low only in very rare and burden or vice versa. circumstances Example from ACP’s guideline on treatment of major depressive disorder: “ACP recommends that clinicians select between either cognitive behavioral therapy or second-generation antidepressants to treat patients with major depressive disorder after discussing treatment effects, adverse effect profiles, cost, accessibility, and preferences with the patient (Grade: strong recommendation, moderate-quality evidence)”.?

Example from ACP’s guideline on treatment of major depressive disorder: “ACP recommends that clinicians select between either cognitive behavioral therapy or second-generation antidepressants to treat patients with major depressive disorder after discussing treatment effects, adverse effect profiles, cost, accessibility, and preferences with the patient (Grade: strong recommendation, moderate-quality evidence)”.? Conditional High, moderate, or low Benefits probably Applies to many Policymaking will require substantial debates and outweigh risks and patients but may involvement of many stakeholders. Policies are also burden, or vice versa, differ depending on more likely to vary between regions. Performance but there is appreciable circumstances or indicators would have to focus on the fact that uncertainty. patients’ values and adequate deliberation about the management preferences. options has taken place. Example from ACP’s guideline on noninvasive treatments of acute, subacute, and chronic low back pain: “In patients with chronic low back pain who have had an inadequate response to nonpharmacologic therapy, clinicians and patients should consider pharmacologic treatment with nonsteroidal anti-inflammatory drugs as first-line therapy, or tramadol or duloxetine as second-line therapy. Clinicians should only consider opioids as an option in patients who have failed the aforementioned treatments and only if the potential benefits outweigh the risks for individual patients and after a discussion of known risks and realistic benefits with patients. (Grade: weak recommendation, moderate-quality evidence)”.

Example from ACP’s guideline on noninvasive treatments of acute, subacute, and chronic low back pain: “In patients with chronic low back pain who have had an inadequate response to nonpharmacologic therapy, clinicians and patients should consider pharmacologic treatment with nonsteroidal anti-inflammatory drugs as first-line therapy, or tramadol or duloxetine as second-line therapy. Clinicians should only consider opioids as an option in patients who have failed the aforementioned treatments and only if the potential benefits outweigh the risks for individual patients and after a discussion of known risks and realistic benefits with patients. (Grade: weak recommendation, moderate-quality evidence)”. FIGURE 2. Grading the certainty of evidence and strength of recommendations for ACP clinical guidelines using GRADE. ACP = American College of Physicians; GRADE = Grading of Recommendations Assessment, Development and Evaluation. *Qaseem A, Barry MJ, Kansagara D; Clinical Guidelines C of the American College of Physicians. Nonpharmacologic versus pharmacologic treatment of adult patients with major depressive disorder: a clinical practice guideline from the American College of Physicians. Ann Intern Med. 2016;164:350-9. [PMID: 26857948] doi:10.7326/M15-2570 *Qaseem A, Wilt TJ, McLean RM, Forciea MA; Clinical Guidelines Committee of the American College of Physicians. Noninvasive treatments for acute, subacute, and chronic low back pain: a clinical practice guideline from the American College of Physicians. Ann Intern Med. 2017;166:514-30. [PMID: 28192789] doi:10.7326/M16-2367 Reproduced with permission from Qaseem A, Kansagara D, Lin JS, et al; Clinical Guidelines Committee of the American College of Physicians. The development of clinical guidelines and guidance statements by the Clinical Guidelines Committee of the American College of Physicians: update of methods. Ann Intern Med. 2019;170:867. [PMID: 31181568] doi:10.7326/M18-3290 ©2019, American College of Physicians.

Reproduced with permission from Qaseem A, Kansagara D, Lin JS, et al; Clinical Guidelines Committee of the American College of Physicians. The development of clinical guidelines and guidance statements by the Clinical Guidelines Committee of the American College of Physicians: update of methods. Ann Intern Med. 2019;170:867. [PMID: 31181568] doi:10.7326/M18-3290 ©2019, American College of Physicians. which all participants receive the treatment being studied. Systematic Reviews and Meta-analyses Because there is no comparison or control group, the ability of Narrative review articles authored by topic experts are a com- the quasi-experimental design to determine cause and effect is mon source of information for clinicians. Although narrative limited. reviews are a valuable learning resource, the information they contain is influenced by the experiences of the authors and are Observational Studies inherently biased. In contrast, systematic reviews involve a In observational studies, no treatment is imposed on the per- systematic search of the literature using predefined criteria to sons being studied. Instead, researchers observe naturally collect all studies that address the topic. Systematic reviews occurring events in individuals who have or have not been should have clear methods and, similar to other scientific exposed to a risk factor or treatment. The relationships studies, should be reproducible. The systematic approach min- between risk factors and disease may be explored, but the abil- imizes selection bias and increases the strength of the infor- ity to assess for causality is limited. mation presented. A systematic review may also include a Common observational study designs include ecologic, meta-analysis if the results of the studies in the review are cross-sectional, case-control, and cohort studies. Case reports combined and included in a statistical analysis. or case series, which describe an individual case or a series of In general, a study with a larger sample size has greater related cases, are not true studies but have the potential to power to discriminate small differences between groups; how- identify new or unique relationships to be further studied. ever, larger studies carry a higher cost and are logistically more Descriptions, uses, and examples of observational study complicated to administer. Because meta-analyses pool data designs are provided in Table 5. from multiple individual studies and increase sample size and

which all participants receive the treatment being studied. Systematic Reviews and Meta-analyses Because there is no comparison or control group, the ability of Narrative review articles authored by topic experts are a com- the quasi-experimental design to determine cause and effect is mon source of information for clinicians. Although narrative limited. reviews are a valuable learning resource, the information they contain is influenced by the experiences of the authors and are Observational Studies inherently biased. In contrast, systematic reviews involve a In observational studies, no treatment is imposed on the per- systematic search of the literature using predefined criteria to sons being studied. Instead, researchers observe naturally collect all studies that address the topic. Systematic reviews occurring events in individuals who have or have not been should have clear methods and, similar to other scientific exposed to a risk factor or treatment. The relationships studies, should be reproducible. The systematic approach min- between risk factors and disease may be explored, but the abil- imizes selection bias and increases the strength of the infor- ity to assess for causality is limited. mation presented. A systematic review may also include a Common observational study designs include ecologic, meta-analysis if the results of the studies in the review are cross-sectional, case-control, and cohort studies. Case reports combined and included in a statistical analysis. or case series, which describe an individual case or a series of In general, a study with a larger sample size has greater related cases, are not true studies but have the potential to power to discriminate small differences between groups; how- identify new or unique relationships to be further studied. ever, larger studies carry a higher cost and are logistically more Descriptions, uses, and examples of observational study complicated to administer. Because meta-analyses pool data designs are provided in Table 5. from multiple individual studies and increase sample size and 5

Clinical Decision Making and Interpreting the Medical Literature TABLE 4. Experimental Study Designs | Study Design Description Example Best Uses and Common Common Measure | \ | Question Type Limitations to Use _ of Association | | * ads | L and Validity

| Question Type Limitations to Use _ of Association | | * ads | L and Validity Quasi-experimental All participants Study of a new Establishing Difficult to clearly Before-and-after | study receive treatment drug for hyper- treatment effect, establish that comparison of | of interest lipidemia in defining risks for treatment caused outcome of interest | whichthe drugis adverse events outcome | given to all study Risk for | participants, lipid confounding, levels are especially for compared before factors not known and after partici- pants receive the drug, and participants are monitored for any adverse effects of the drug

Quasi-experimental All participants Study of a new Establishing Difficult to clearly Before-and-after | study receive treatment drug for hyper- treatment effect, establish that comparison of | of interest lipidemia in defining risks for treatment caused outcome of interest | whichthe drugis adverse events outcome | given to all study Risk for | participants, lipid confounding, levels are especially for compared before factors not known and after partici- pants receive the drug, and participants are monitored for any adverse effects of the drug Nonrandomized Treatment is Study of surgical When blinding to Risk for Difference between controlled trial allocated ina and nonsurgical treatment type is confounding, treatment groups in prespecified treatment options not practical or especially for outcome of interest manner, but in which enrolled possible factors not known Measure reported not using participants are When patients/ Risk for selection varies depending on randomization; assigned to the participants may bias if either outcome of interest comparison is treatment type not be amenable to patients or and statistical made between they prefer to being randomly researchers are technique different treatment receive assigned to a allowed a choice in groups or between treatment or when treatment allocated treatment and randomization is control groups not ethically permitted

Nonrandomized Treatment is Study of surgical When blinding to Risk for Difference between controlled trial allocated ina and nonsurgical treatment type is confounding, treatment groups in prespecified treatment options not practical or especially for outcome of interest manner, but in which enrolled possible factors not known Measure reported not using participants are When patients/ Risk for selection varies depending on randomization; assigned to the participants may bias if either outcome of interest comparison is treatment type not be amenable to patients or and statistical made between they prefer to being randomly researchers are technique different treatment receive assigned to a allowed a choice in groups or between treatment or when treatment allocated treatment and randomization is control groups not ethically permitted Cluster randomized Participants are Study of effect Treatments for Complicated Difference between trial randomly assigned of diabetes which randomiza- statistical analysis treatment groups in as a unit instead of education on tion of individual outcome of interest May be difficult to individually average patients is not ensure groups Measure reported hemoglobin A,, feasible or ethical being compared varies depending on value, in which When effect at a are similar (risk for outcome of interest participants are level larger than an confounding) and statistical community individual level (i.e., technique primary care institution or health practices system) is being studied

Cluster randomized Participants are Study of effect Treatments for Complicated Difference between trial randomly assigned of diabetes which randomiza- statistical analysis treatment groups in as a unit instead of education on tion of individual outcome of interest May be difficult to individually average patients is not ensure groups Measure reported hemoglobin A,, feasible or ethical being compared varies depending on value, in which When effect at a are similar (risk for outcome of interest participants are level larger than an confounding) and statistical community individual level (i.e., technique primary care institution or health practices system) is being studied | Randomized Participants are Study comparing Considered to be Generalizability is Difference between | controlled trial randomly assigned a new class of the highest level limited to the treatment groups in to one of the lipid-lowering of evidence for population outcome of interest treatment options drugs versus treatment, etiology, included in the trial Measure reported (including control/ statin therapy in and prevention Expensive and varies depending on placebo group) patients with no studies resource intensive outcome of interest known heart Confounding is to conduct and statistical disease but with minimized if study technique predicted 10-year is large enough cardiovascular risk higher than 20%, in which participants are randomly assigned to one of the two treatments, pills are visually identical, and investigators and statisticians do not know to which treatment participants are assigned

Clinical Decision Making and Interpreting the Medical Literature TABLE 5. Observational Study Designs | Study Description Example Best Uses and Common Common Measure | Design Clinical Question Limitations to Use _ of Association Type and Validity

| Study Description Example Best Uses and Common Common Measure | Design Clinical Question Limitations to Use _ of Association Type and Validity Cohort study _ Participants are Prospective cohort: May provide the Risk for recall bias Risk ratio (or divided into groups Women in the first 5 years highest level of with retrospective relative risk): risk for | based on the after menopause onset evidence for cohort design developing the presence or who are receiving diagnosis or outcome being Risk for participants absence of a critical estrogen therapy are prognosis studies studied if exposed to be lost to follow- exposure and are compared with women to the risk factor of | Useful for studying up in prospective evaluated for not receiving estrogen interest compared rare exposures, studies with differences therapy. All are assessed with those who outcomes that lengthy follow-up between the yearly over 10 years of have not been develop over time, intervals groups in the follow-up for the exposed and survival (time- development of a development of Can be lengthy and to-event analyses) disease or outcome cognitive impairment expensive to follow Provides true participants until Participants may be Retrospective cohort: incidence of they develop recruited before the Postmenopausal women disease disease outcome of interest who received estrogen develops and replacement therapy in Difficult to match followed forward in the first 5 years after both cohorts in all time (prospective menopause onset are variables except for cohort) or recruited compared with those exposure and grouped who did not receive according to estrogen with respect exposure at atime to development of when the outcome dementia. Women who would have already experienced menopause occurred 10 or more years ago are (retrospective included in the study

Cohort study _ Participants are Prospective cohort: May provide the Risk for recall bias Risk ratio (or divided into groups Women in the first 5 years highest level of with retrospective relative risk): risk for | based on the after menopause onset evidence for cohort design developing the presence or who are receiving diagnosis or outcome being Risk for participants absence of a critical estrogen therapy are prognosis studies studied if exposed to be lost to follow- exposure and are compared with women to the risk factor of | Useful for studying up in prospective evaluated for not receiving estrogen interest compared rare exposures, studies with differences therapy. All are assessed with those who outcomes that lengthy follow-up between the yearly over 10 years of have not been develop over time, intervals groups in the follow-up for the exposed and survival (time- development of a development of Can be lengthy and to-event analyses) disease or outcome cognitive impairment expensive to follow Provides true participants until Participants may be Retrospective cohort: incidence of they develop recruited before the Postmenopausal women disease disease outcome of interest who received estrogen develops and replacement therapy in Difficult to match followed forward in the first 5 years after both cohorts in all time (prospective menopause onset are variables except for cohort) or recruited compared with those exposure and grouped who did not receive according to estrogen with respect exposure at atime to development of when the outcome dementia. Women who would have already experienced menopause occurred 10 or more years ago are (retrospective included in the study | cohort)

Cohort study _ Participants are Prospective cohort: May provide the Risk for recall bias Risk ratio (or divided into groups Women in the first 5 years highest level of with retrospective relative risk): risk for | based on the after menopause onset evidence for cohort design developing the presence or who are receiving diagnosis or outcome being Risk for participants absence of a critical estrogen therapy are prognosis studies studied if exposed to be lost to follow- exposure and are compared with women to the risk factor of | Useful for studying up in prospective evaluated for not receiving estrogen interest compared rare exposures, studies with differences therapy. All are assessed with those who outcomes that lengthy follow-up between the yearly over 10 years of have not been develop over time, intervals groups in the follow-up for the exposed and survival (time- development of a development of Can be lengthy and to-event analyses) disease or outcome cognitive impairment expensive to follow Provides true participants until Participants may be Retrospective cohort: incidence of they develop recruited before the Postmenopausal women disease disease outcome of interest who received estrogen develops and replacement therapy in Difficult to match followed forward in the first 5 years after both cohorts in all time (prospective menopause onset are variables except for cohort) or recruited compared with those exposure and grouped who did not receive according to estrogen with respect exposure at atime to development of when the outcome dementia. Women who would have already experienced menopause occurred 10 or more years ago are (retrospective included in the study | cohort) Case-control Participants are Mothers of infants with May be the only Risk for recall bias, Odds ratio: odds of | study divided into groups and without a rare way to collect because events having been exposed _ | based on the congenital heart disease enough data to being studied have tothe risk factor of | presence or are compared with meaningfully already occurred interest in those absence of a respect to prenatal assess relationships and participants with the condition disease or exposure to medications about exposures in with disease are being studied condition and suspected of being rare diseases more likely than versus those who compared with associated with the controls to recall do not have the Can evaluate respect to past disease past exposures condition multiple exposures exposures Difficult to match controls to cases

Case-control Participants are Mothers of infants with May be the only Risk for recall bias, Odds ratio: odds of | study divided into groups and without a rare way to collect because events having been exposed _ | based on the congenital heart disease enough data to being studied have tothe risk factor of | presence or are compared with meaningfully already occurred interest in those absence of a respect to prenatal assess relationships and participants with the condition disease or exposure to medications about exposures in with disease are being studied condition and suspected of being rare diseases more likely than versus those who compared with associated with the controls to recall do not have the Can evaluate respect to past disease past exposures condition multiple exposures exposures Difficult to match controls to cases Cannot determine the true incidence of disease

Case-control Participants are Mothers of infants with May be the only Risk for recall bias, Odds ratio: odds of | study divided into groups and without a rare way to collect because events having been exposed _ | based on the congenital heart disease enough data to being studied have tothe risk factor of | presence or are compared with meaningfully already occurred interest in those absence of a respect to prenatal assess relationships and participants with the condition disease or exposure to medications about exposures in with disease are being studied condition and suspected of being rare diseases more likely than versus those who compared with associated with the controls to recall do not have the Can evaluate respect to past disease past exposures condition multiple exposures exposures Difficult to match controls to cases Cannot determine the true incidence of disease Cross- Evaluates the Surveys conducted by Best used to Cannot be used to Prevalence ratio sectional relationship health services groups determine disease determine cause (prevalence odds study between exposures may serve as the basis for prevalence and and effect because ratio): prevalence and health these studies. Examples generate temporal sequence of the outcome or outcomes ina include the National hypotheses is unknown and condition in population of Health and Nutrition comparison groups individuals with the Can also assess interest Examination Survey and are not used exposure or risk the relatedness the Behavioral Risk Factor factor of interest Factors and between two Surveillance Survey, both compared with outcomes are exposures of conducted by the CDC prevalence in measured ata interest individuals without single point in time the risk factor

Cross- Evaluates the Surveys conducted by Best used to Cannot be used to Prevalence ratio sectional relationship health services groups determine disease determine cause (prevalence odds study between exposures may serve as the basis for prevalence and and effect because ratio): prevalence and health these studies. Examples generate temporal sequence of the outcome or outcomes ina include the National hypotheses is unknown and condition in population of Health and Nutrition comparison groups individuals with the Can also assess interest Examination Survey and are not used exposure or risk the relatedness the Behavioral Risk Factor factor of interest Factors and between two Surveillance Survey, both compared with outcomes are exposures of conducted by the CDC prevalence in measured ata interest individuals without single point in time the risk factor Ecological Evaluates the Comparison of rates of Used to identify Cannot be used to Rate ratio: study prevalence or colon cancer in countries compelling trends determine cause prevalence of a extent of an with higher versus lower for an exposure of and effect condition in groups | exposure across rates of red meat interest across with higher rates of distinct populations Incorrect consumption different groups, exposure compared | assignment of which can be to prevalence in causation based on investigated groups with lower | ecologic study subsequently in rates of exposure results is termed the more focused ecological fallacy studies

Clinical Decision Making and Interpreting the Medical Literature statistical power, the ability to detect a difference between the confidence interval range. Narrower ranges imply greater confi- studied groups increases. Meta-analysis of high-quality ran- dence, or certainty, that the reported value reflects the true value. domized controlled trials is considered one of the highest Importantly, statistical significance is not equivalent to levels of clinical evidence. Factors that limit the ability to com- clinical significance. A treatment may result in a statistically bine study results meaningfully include differences in the significant decrease in symptoms, for example, but these study populations, research measures, and statistical tech- changes may not be clinically important. niques; these differences are termed heterogeneity. The quality of a systematic review can be evaluated against best practice e Experimental studies systematically expose some or all criteria, such as the Preferred Reporting Items for Systematic of the individuals in the study to a potential causative or Reviews and Meta-Analyses (PRISMA) criteria. protective factor; the strongest experimental design is Statistical Analysis the randomized controlled trial.

statistical power, the ability to detect a difference between the confidence interval range. Narrower ranges imply greater confi- studied groups increases. Meta-analysis of high-quality ran- dence, or certainty, that the reported value reflects the true value. domized controlled trials is considered one of the highest Importantly, statistical significance is not equivalent to levels of clinical evidence. Factors that limit the ability to com- clinical significance. A treatment may result in a statistically bine study results meaningfully include differences in the significant decrease in symptoms, for example, but these study populations, research measures, and statistical tech- changes may not be clinically important. niques; these differences are termed heterogeneity. The quality of a systematic review can be evaluated against best practice e Experimental studies systematically expose some or all criteria, such as the Preferred Reporting Items for Systematic of the individuals in the study to a potential causative or Reviews and Meta-Analyses (PRISMA) criteria. protective factor; the strongest experimental design is Statistical Analysis the randomized controlled trial. Validity is the degree to which the study results reflect what is e In observational studies, researchers observe naturally correct or true. It is distinct from reliability, a construct that occurring events in individuals who have or have not measures how consistent the results are upon repeated meas- been exposed to a risk factor or treatment; relationships urement. Validity can be further categorized as internal or between risk factors and disease may be explored, but external. Internal validity refers to how well study error is mini- the ability to assess for causality is limited. mized, whereas external validity is the extent to which the study results can be applied beyond the study setting, or the generaliz- Application of Study Results ability. Threats to validity include systematic and random error. Concepts rooted in evidence-based medicine, including sensi- Systematic error is the influence of confounders and bias tivity, specificity, predictive values, likelihood ratios, risk on study results. Confounders are factors other than the vari- reduction measures, and numbers needed to treat and harm, ables being studied that are associated with the studied popu- can be helpful in applying findings from the literature to lation and may affect the end point being assessed. A patient care (Table 6 and Table 7). Although useful, they are cross-sectional study, for example, may unexpectedly show also subject to specific shortcomings (Table 8). that adults receiving statin therapy have increased risk for cardiovascular events compared with those not taking statin therapy. However, the presence of a powerful confounder, Sensitivity, Specificity, and Predictive Values specifically that patients receive statin therapy because they Sensitivity and specificity reflect the accuracy of a diagnostic or

Validity is the degree to which the study results reflect what is e In observational studies, researchers observe naturally correct or true. It is distinct from reliability, a construct that occurring events in individuals who have or have not measures how consistent the results are upon repeated meas- been exposed to a risk factor or treatment; relationships urement. Validity can be further categorized as internal or between risk factors and disease may be explored, but external. Internal validity refers to how well study error is mini- the ability to assess for causality is limited. mized, whereas external validity is the extent to which the study results can be applied beyond the study setting, or the generaliz- Application of Study Results ability. Threats to validity include systematic and random error. Concepts rooted in evidence-based medicine, including sensi- Systematic error is the influence of confounders and bias tivity, specificity, predictive values, likelihood ratios, risk on study results. Confounders are factors other than the vari- reduction measures, and numbers needed to treat and harm, ables being studied that are associated with the studied popu- can be helpful in applying findings from the literature to lation and may affect the end point being assessed. A patient care (Table 6 and Table 7). Although useful, they are cross-sectional study, for example, may unexpectedly show also subject to specific shortcomings (Table 8). that adults receiving statin therapy have increased risk for cardiovascular events compared with those not taking statin therapy. However, the presence of a powerful confounder, Sensitivity, Specificity, and Predictive Values specifically that patients receive statin therapy because they Sensitivity and specificity reflect the accuracy of a diagnostic or are at increased risk for cardiovascular events, influences the screening test (see Table 6). Sensitivity is the ability of the test to outcome of the study. The impact of confounding can be detect disease in those who truly have it, or the probability that

Validity is the degree to which the study results reflect what is e In observational studies, researchers observe naturally correct or true. It is distinct from reliability, a construct that occurring events in individuals who have or have not measures how consistent the results are upon repeated meas- been exposed to a risk factor or treatment; relationships urement. Validity can be further categorized as internal or between risk factors and disease may be explored, but external. Internal validity refers to how well study error is mini- the ability to assess for causality is limited. mized, whereas external validity is the extent to which the study results can be applied beyond the study setting, or the generaliz- Application of Study Results ability. Threats to validity include systematic and random error. Concepts rooted in evidence-based medicine, including sensi- Systematic error is the influence of confounders and bias tivity, specificity, predictive values, likelihood ratios, risk on study results. Confounders are factors other than the vari- reduction measures, and numbers needed to treat and harm, ables being studied that are associated with the studied popu- can be helpful in applying findings from the literature to lation and may affect the end point being assessed. A patient care (Table 6 and Table 7). Although useful, they are cross-sectional study, for example, may unexpectedly show also subject to specific shortcomings (Table 8). that adults receiving statin therapy have increased risk for cardiovascular events compared with those not taking statin therapy. However, the presence of a powerful confounder, Sensitivity, Specificity, and Predictive Values specifically that patients receive statin therapy because they Sensitivity and specificity reflect the accuracy of a diagnostic or are at increased risk for cardiovascular events, influences the screening test (see Table 6). Sensitivity is the ability of the test to outcome of the study. The impact of confounding can be detect disease in those who truly have it, or the probability that reduced by design strategies (e.g., matching) and evaluation the test result will be positive in a patient with the disease.

are at increased risk for cardiovascular events, influences the screening test (see Table 6). Sensitivity is the ability of the test to outcome of the study. The impact of confounding can be detect disease in those who truly have it, or the probability that reduced by design strategies (e.g., matching) and evaluation the test result will be positive in a patient with the disease. techniques (e.g., stratification of the analysis and regression Specificity is the ability of the test to correctly identify those with- techniques). The strongest protection against confounding is out the disease, or the probability that the test result will be nega- provided by random assignment of the study intervention and tive in a patient without the disease. Therefore, sensitive tests are blinding of the participants, study investigators, and individu- those with minimal rates of false-negative results, and specific als analyzing the study results. Bias refers to the presence of tests are those with minimal rates of false-positive results. factors that skew study results in a specific direction. Bias can Positive and negative predictive values (see Table 6) reflect be introduced at any point in a study design, including patient the validity of a positive or negative test result for predicting recruitment and outcome measurement. Careful study design the presence or absence of disease in a specific population. can decrease the likelihood of bias; however, once bias is pre- Predictive values are determined not only by sensitivity and sent, it cannot be eliminated or minimized. specificity but also by disease prevalence.

recruitment and outcome measurement. Careful study design the presence or absence of disease in a specific population. can decrease the likelihood of bias; however, once bias is pre- Predictive values are determined not only by sensitivity and sent, it cannot be eliminated or minimized. specificity but also by disease prevalence. Random error is introduced into studies by random vari- ability or chance. The presence of random error can be reduced Likelihood Ratios by increasing study size and using precise measurement strate- Likelihood ratios (LRs) provide a measure of the strength of asso- gies. Statistical significance, which is reported using P values or ciation betweena clinical finding and a specific disease; in effect, confidence intervals, helps quantify how likely the results of a they translate the impact of sensitivity and specificity on the study are due to random error. A P value less than or equal to disease probability. Each clinical finding, including symptoms, 0.05 (<5% probability that chance alone accounts for the result) physical findings, and laboratory and imaging results, has a posi- is often considered to indicate a statistically significant finding. tive and negative LR for the diseases with which they are associ- However, this determination is arbitrary, and some researchers ated. The positive LR is used when the test result is positive, and may opt for a different level (typically more stringent). the negative LR is used when the test result is negative. Confidence intervals express the range of values within LRs are used in conjunction with the pretest probability which the true result falls; the 95% confidence interval means of disease (the probability of disease in the patient before that there is 95% confidence that the true value is within the the test result is considered). The interaction of the pretest

Clinical Decision Making and Interpreting the Medical Literature TABLE 6. Common Terms Used in Interpretation of the Medical Literature for Diagnostic Tests Term Definition Calculation Notes Prevalence (Prev) Proportion of patients with the Prev=(TP+FN)(TP+FP+ disease in the population FN +TN) Sensitivity (Sn) Proportion of patients with the Sn=TP/(TP + FN) disease who have a positive test result Specificity (Sp) Proportion of patients without the Sp =TN/AFP +TN) disease who have a negative test result Positive predictive value (PPV) Proportion of patients with a PPV=TPA(TP + FP) Increases with increasing positive test result who have the prevalence disease Negative predictive value Proportion of patients with a NPV=TNATN + FN) Increases with decreasing (NPV) negative test result who do not prevalence have the disease Positive likelihood ratio (LR+) Ratio of the probability of a LR+=Sn/(1 - Sp) | positive test result among patients with the disease to the probability of a positive result among patients without the disease

Positive likelihood ratio (LR+) Ratio of the probability of a LR+=Sn/(1 - Sp) | positive test result among patients with the disease to the probability of a positive result among patients without the disease Negative likelihood ratio (LR—-) Ratio of the probability of a LR-=(1-Sn)/Sp negative test result among patients with the disease to the probability of a negative result in patients without the disease Pretest odds Odds that a patient has the Pretest odds = pretest | disease before the test is probability/(1 — pretest | performed probability) Posttest odds Odds that a patient has the Posttest odds = pretest LR+ is used if result of test is disease after a test is performed odds x LR positive; LR— is used if result of test is negative

Posttest odds Odds that a patient has the Posttest odds = pretest LR+ is used if result of test is disease after a test is performed odds x LR positive; LR— is used if result of test is negative Anomogram is available to calculate posttest probability using pretest probability and LR without having to convert pretest probability to odds (see Figure 3) Pretest probability Proportion of patients with the Pretest probability can be | | disease before a test is performed estimated from population | | prevalence, clinical risk | | | calculators, or clinical } } experience if no evidence- based tools exist Posttest probability Proportion of patients with the Posttest probability = disease after a test is performed posttest odds/(1 + posttest odds)

Posttest probability Proportion of patients with the Posttest probability = disease after a test is performed posttest odds/(1 + posttest odds) FN = false negative; FP = false positive; TN = true negative; TP = true positive. probability and LR results in a posttest probability of disease. disease probability by 15%, 30%, and 45%, respectively; nega- The posttest probability may then serve as the rationale for tive LR values of 0.5, 0.2, and 0.1 correspond to a decrease in treating a patient for the disease (if the positive test resulted in disease probability by 15%, 30%, and 45%, respectively. Larger a high posttest probability) or ruling out a diagnosis (if the positive LRs and smaller negative LRs are more apt to affect negative test resulted in a very low posttest probability). clinical decisions. Equations for converting pretest probability (a percentage) to Clinical tools for calculating LRs and their effect on prob- pretest odds (a ratio), applying the LR, and converting posttest ability of disease, such as one from the Centre for Evidence- odds to posttest probability are available in Table 6. Based Medicine (https://www.cebm.ox.ac.uk/resources/ For ease of clinical use, several general LR rules apply. ebm-tools/catmaker-and-ebm-calculators), are widely avail- Positive LR values of 2, 5, and 10 correspond to an increase in able. Alternatively, nomograms may be used to extrapolate the 9

Clinical Decision Making and Interpreting the Medical Literature TABLE 7. Common Terms Used in Interpretation of the Medical Literature for Therapeutics Term Definition Calculation Notes \ Absolute risk (AR) Probability of an event AR = patients with event in Also known as event rate; can | occurring in a group during a group/total patients in group be for benefits or harms. Often, specified period an experimental event rate (EER) is compared with a control event rate (CER) Relative risk (RR) Ratio of the probability of RR=EER/CER Used in cohort studies and developing a disease with a randomized controlled trials risk factor present to the probability of developing the disease without the risk factor present Absolute risk reduction (ARR) Difference in rates of events ARR=|EER-CER| between the EER and the CER Relative risk reduction (RRR) Ratio of absolute risk reduction RRR =| EER — CER |/CER to the event rate among controls

Absolute risk reduction (ARR) Difference in rates of events ARR=|EER-CER| between the EER and the CER Relative risk reduction (RRR) Ratio of absolute risk reduction RRR =| EER — CER |/CER to the event rate among controls Number needed to treat (NNT) Number of patients needed to NNT =1/ARR Agood estimate of the effect receive a treatment for one size additional patient to benefit Number needed to harm Number of patients needed to NNH = 1/ARI ARI is the absolute risk increase (NNH) receive a treatment for one and equals | EER —- CER| when additional patient to be harmed the event is an unfavorable outcome (e.g., drug side effect) | TABLE 8. Application of Evidence-Based Medicine Concepts Evidence-Based Medicine Shortcomings and Examples Concepts Sensitivity, specificity, and Predictive values are often used in preference to sensitivity and specificity in clinical practice; however, predictive values if disease prevalence in the study population is different than in the clinical population, the predictive values will change Example: Screening for cervical cancer with HPV testing alone in women aged 21-29 years results in a high false-positive rate compared with testing in women aged 30-65 years owing to higher rates of HPV exposure Likelihood ratio Likelihood ratios are not interchangeable with odds ratios

Example: Screening for cervical cancer with HPV testing alone in women aged 21-29 years results in a high false-positive rate compared with testing in women aged 30-65 years owing to higher rates of HPV exposure Likelihood ratio Likelihood ratios are not interchangeable with odds ratios Positive likelihood ratios are not the reciprocals of negative likelihood ratios Example: Dry axillae are supportive of hypovolemia (LR+, 2.8), but their absence does not exclude hypovolemia (LR-, 0.6). Absolute and relative risk Absolute and relative risk reduction calculations do not identify whether the difference is clinically reduction significant. A large relative risk may distract from small absolute changes Example: Ezetimibe reduces stroke risk from 32/1000 to 27/1000. Relative risk reduction is 17%, but absolute risk reduction is 0.5%. Number needed to treat or These measures do not qualitatively compare the magnitude of the benefit or harm number needed to harm Example: An emergency department visit for a nosebleed requiring packing may have fewer significant long-term consequences than stroke in patients with atrial fibrillation External validity Exclusion criteria or bias in recruitment may result in a study population that is not representative of the (generalizability) general population Example: The reduction in risk for myocardial infarction associated with a drug studied in a secondary prevention population cannot be applied to a primary prevention population Internal validity and bias Intention-to-treat and per-protocol analyses can have significantly different results

Example: The reduction in risk for myocardial infarction associated with a drug studied in a secondary prevention population cannot be applied to a primary prevention population Internal validity and bias Intention-to-treat and per-protocol analyses can have significantly different results Example: A high rate of drug side effects may lead to frequent discontinuation, reducing efficacy in an intention-to-treat analysis. A per-protocol analysis would exclude those patients and potentially overestimate the benefit in routine clinical practice HPV = human papillomavirus; LR+ = positive likelihood ratio; LR- = negative likelihood ratio. — 10

Medical Ethics and Professionalism 014- - 99 rate of acute coronary syndrome in 5 years is 3.6% in the treated group and 4.8% in the control group. The absolute risk 0.2—4- + 98 reduction is 1.2%. The risk ratio is 75%, and the relative risk reduction is 25%. Although the absolute difference in rates 0.5—— + 95 between the two groups is small, the effect may appear sub- 2000 + stantial when relative risk reduction is reported. i= 1000 a a 90

014- - 99 rate of acute coronary syndrome in 5 years is 3.6% in the treated group and 4.8% in the control group. The absolute risk 0.2—4- + 98 reduction is 1.2%. The risk ratio is 75%, and the relative risk reduction is 25%. Although the absolute difference in rates 0.5—— + 95 between the two groups is small, the effect may appear sub- 2000 + stantial when relative risk reduction is reported. i= 1000 a a 90 500 = 2 200 ++ Numbers Needed to Treat and Harm - 80 100-- 7 - The number needed to treat (NNT) or harm (NNH) expresses sal 504 the clinical impact of an intervention. The NNT is the inverse 20+ = 60 of the absolute risk reduction. It represents the number of 10 + 50 10—- patients who must receive a treatment to cause one additional : s+ —+ 40 patient to benefit. The NNT provides a quantifiable measure of 20+ 2 + 30 _ the treatment effect that is easily understood by physicians

500 = 2 200 ++ Numbers Needed to Treat and Harm - 80 100-- 7 - The number needed to treat (NNT) or harm (NNH) expresses sal 504 the clinical impact of an intervention. The NNT is the inverse 20+ = 60 of the absolute risk reduction. It represents the number of 10 + 50 10—- patients who must receive a treatment to cause one additional : s+ —+ 40 patient to benefit. The NNT provides a quantifiable measure of 20+ 2 + 30 _ the treatment effect that is easily understood by physicians 30-- io Se and patients. The acceptability of the NNT depends on the risks associated with the condition, the cost and side effects of 40- + 0.2 50+ = 0.1 i the treatment, and other treatments available. The NNH can be 60 —- 0.05 calculated for studies measuring adverse outcomes but also 1; 70 “ — 0.02 needs to be considered in the context of the benefits of the aA + 0.01 treatment and the severity of the harm incurred. —r 0.005 +2

30-- io Se and patients. The acceptability of the NNT depends on the risks associated with the condition, the cost and side effects of 40- + 0.2 50+ = 0.1 i the treatment, and other treatments available. The NNH can be 60 —- 0.05 calculated for studies measuring adverse outcomes but also 1; 70 “ — 0.02 needs to be considered in the context of the benefits of the aA + 0.01 treatment and the severity of the harm incurred. —r 0.005 +2 + 0.002 elli + 0.001 a1 e Sensitivity is the ability of the test to detect those who L + 0.0005 954 truly have a disease; specificity is the ability of the test + 0.5 to correctly identify those without the disease. 98 _ + 0.2 e Likelihood ratios provide a measure of the strength of association betweena clinical finding and a specific 99 L +o1 disease.

98 _ + 0.2 e Likelihood ratios provide a measure of the strength of association betweena clinical finding and a specific 99 L +o1 disease. e Absolute risk reduction is the difference in the response Pretest Likelihood Posttest to treatment between experimental and control groups. probability ratio probability (%) (%) e The number needed to treat represents the number of patients who must receive a treatment to cause one FIGURE 3. Nomogram for interpreting diagnostic test results. In this nomogram, a straight line drawn from a patient's pretest probability of disease additional patient to benefit. (which is estimated from experience, local data, or published literature) through the likelihood ratio for the test result will indicate the posttest probability of disease. Reproduced with permission from Fagan TJ. Letter: nomogram for Bayes theorem. N EnglJ Med. 1975;293:257. Medical Ethics and Professionalism [PMID: 1143310] ©1975, Massachusetts Medical Society.

e Absolute risk reduction is the difference in the response Pretest Likelihood Posttest to treatment between experimental and control groups. probability ratio probability (%) (%) e The number needed to treat represents the number of patients who must receive a treatment to cause one FIGURE 3. Nomogram for interpreting diagnostic test results. In this nomogram, a straight line drawn from a patient's pretest probability of disease additional patient to benefit. (which is estimated from experience, local data, or published literature) through the likelihood ratio for the test result will indicate the posttest probability of disease. Reproduced with permission from Fagan TJ. Letter: nomogram for Bayes theorem. N EnglJ Med. 1975;293:257. Medical Ethics and Professionalism [PMID: 1143310] ©1975, Massachusetts Medical Society. posttest probability of disease from the pretest probability of Principles of Ethics and disease on the basis of the LR (Figure 3). Professionalism Absolute and Relative Risk Reduction Physicians should be guided in their practice by the four basic Absolute risk reduction and relative risk reduction are meas- principles of medical ethics, which work in unison to define ures commonly used to report the efficacy of an intervention the duties of the physician to the patient (Table 9). In applying (see Table 7). Absolute risk reduction is the difference in the these principles, physicians must consider the specific clinical response to treatment between experimental and control scenario, including cultural and patient-specific factors, to groups. Absolute risk reduction is particularly relevant when determine an appropriate course of action. At times, however, the measure reported is a clinically meaningful end point (e.g., ethical principles may seem in conflict, leading to ethical the incidence of cardiovascular events in studies of lipid-low- dilemmas. For example, respecting the autonomy of a compe- ering therapy). Because these differences may seem small, tent patient who is making poor decisions may seem to con- investigators may also report the relative risk or risk ratio as tradict the principles of beneficence and nonmaleficence. well as relative risk reduction. Ethical dilemma resolution can often be accomplished by The difference in the magnitude of these measures can be (1) evaluating the medical situation; (2) framing the ethical illustrated with a study of lipid-lowering therapy in which the question or dilemma and considering all available options;

posttest probability of disease from the pretest probability of Principles of Ethics and disease on the basis of the LR (Figure 3). Professionalism Absolute and Relative Risk Reduction Physicians should be guided in their practice by the four basic Absolute risk reduction and relative risk reduction are meas- principles of medical ethics, which work in unison to define ures commonly used to report the efficacy of an intervention the duties of the physician to the patient (Table 9). In applying (see Table 7). Absolute risk reduction is the difference in the these principles, physicians must consider the specific clinical response to treatment between experimental and control scenario, including cultural and patient-specific factors, to groups. Absolute risk reduction is particularly relevant when determine an appropriate course of action. At times, however, the measure reported is a clinically meaningful end point (e.g., ethical principles may seem in conflict, leading to ethical the incidence of cardiovascular events in studies of lipid-low- dilemmas. For example, respecting the autonomy of a compe- ering therapy). Because these differences may seem small, tent patient who is making poor decisions may seem to con- investigators may also report the relative risk or risk ratio as tradict the principles of beneficence and nonmaleficence. well as relative risk reduction. Ethical dilemma resolution can often be accomplished by The difference in the magnitude of these measures can be (1) evaluating the medical situation; (2) framing the ethical illustrated with a study of lipid-lowering therapy in which the question or dilemma and considering all available options; 11