Browse the corpus

General Internal Medicine a a a B I o stat I st I c5 Sensitivity, Specificity, Predictive Values, and Likelihood Ratios Sensitivity is the ability of a test to detect a disease when it is truly present. Specificity is the ability of a test to exclude disease when it is truly absent. The positive predictive value is the probability that a positive test is a true positive result' and the negative predictive value is the probability that a negative test is a true negative result. DOT'T BE TRICKED . As the prevalence of a condition increases, the positive predictive value increases and the negative predictive value decreases. . Changes in prevalence do not alter the sensitivity or speciflcity but do alter the predictive values. A likelihood ratio (LR) is a measurement of the odds of having a disease independent of the disease prevalence. Separate LRs are calculated for use when a test result is positive (LR+) or negative (LR-): o LR+ = Sensitivity/(1 Specificity) o LR = (1 - SensitiviB")/Specificity . LR+ of 2, 5, and 10 increase the probability of disease by approximately 15%, 3o%, and 45%, respectively. . LR- of 0.5, 0.2, and 0.1 decrease the probability of disease by approximately 15%, 30%, and 45%, respectively

DOT'T BE TRICKED . As the prevalence of a condition increases, the positive predictive value increases and the negative predictive value decreases. . Changes in prevalence do not alter the sensitivity or speciflcity but do alter the predictive values. A likelihood ratio (LR) is a measurement of the odds of having a disease independent of the disease prevalence. Separate LRs are calculated for use when a test result is positive (LR+) or negative (LR-): o LR+ = Sensitivity/(1 Specificity) o LR = (1 - SensitiviB")/Specificity . LR+ of 2, 5, and 10 increase the probability of disease by approximately 15%, 3o%, and 45%, respectively. . LR- of 0.5, 0.2, and 0.1 decrease the probability of disease by approximately 15%, 30%, and 45%, respectively Study Designs gTUDY TABLE: Characteristics of Study Designs TyPe Characteristics Cross-sectional The presence of the presumed risk factor and presence of the outcome are measured at one point in time in a population. Retrospective (case Subjects are divided into groups based on the presence or absence ofthe outcome of interest, and then the control) frequency of risk factors in each group is compared. Prospective (cohort) Su bjects are divided into groups based on the presence or absence of the presumed risk factor and followed fora period of time. At the end of the study, the frequency of the outcome is compared. Randomized Subjects are randomly divided into groups; one group receives the intervention (patients and researchers may controlled trial be blinded to treatment, termed double-blind) and followed forward in time. At the end of the study, the freq uency of the outcome is com pared. This study design red uces the effect o{ u nmeasu red (confounding)variables that may influence outcomes of a study.

Study Designs gTUDY TABLE: Characteristics of Study Designs TyPe Characteristics Cross-sectional The presence of the presumed risk factor and presence of the outcome are measured at one point in time in a population. Retrospective (case Subjects are divided into groups based on the presence or absence ofthe outcome of interest, and then the control) frequency of risk factors in each group is compared. Prospective (cohort) Su bjects are divided into groups based on the presence or absence of the presumed risk factor and followed fora period of time. At the end of the study, the frequency of the outcome is compared. Randomized Subjects are randomly divided into groups; one group receives the intervention (patients and researchers may controlled trial be blinded to treatment, termed double-blind) and followed forward in time. At the end of the study, the freq uency of the outcome is com pared. This study design red uces the effect o{ u nmeasu red (confounding)variables that may influence outcomes of a study. Systematic review Systematic reviews involve a systematic search of the literature using predefined criteria to collect all studies that address the topic. This systematic approach minimizes selection bias and increases the strength of the information presented. Meta analysis Usually, multiple clinicaltrials identified by a systematic review using similar randomization techniques and interventions can be combined into one large analysis to address very precise clinical questions. The results may be analyzed using the technique of meta-analysis, in which alltrial results are combined to create a single point estimate. By combining the results o{ many trials, the power to detect differences between the intervention and the control or comparator increases. 114

General lnternal Medicine STUDY TABLE: Strength of Research Designs in Descending Order (Strongest to Weakest) Description RCT including systematic reviews of RCTs Clinical trial without randomization Cross-sectional, case-control, or cohort study Evidence based on experience, descriptive studies, or expert opinion Risk Estimates STUDY TABLE: Common Calculations Used in Clinical Research Term Definition Calculation Notes Absolute risk (AR) The probability of an event trft = p3tients with event / total Also known as event rate. occurring in a group during a patients Often, an experimental event specified time period rate (EER) is compared with a control event rate (CER) Relative risk (RR) The ratio of the probability of RR=EER/CER Used in cohort studies and developing a disease with a RCTs risk factor present to the probability of developing the disease with the risk factor absent Absolute risk reduction (ARR) The abso/ute difference in rates ARR=IEER-CERI Critical to understanding of events between an number needed to treat experimental group and a (below) control group Relative risk reduction (RRR) The ratio of ARR to the event ppp=IEER-CERI/CER For very infrequent events, RR rate among controls can be large while AR is small

Risk Estimates STUDY TABLE: Common Calculations Used in Clinical Research Term Definition Calculation Notes Absolute risk (AR) The probability of an event trft = p3tients with event / total Also known as event rate. occurring in a group during a patients Often, an experimental event specified time period rate (EER) is compared with a control event rate (CER) Relative risk (RR) The ratio of the probability of RR=EER/CER Used in cohort studies and developing a disease with a RCTs risk factor present to the probability of developing the disease with the risk factor absent Absolute risk reduction (ARR) The abso/ute difference in rates ARR=IEER-CERI Critical to understanding of events between an number needed to treat experimental group and a (below) control group Relative risk reduction (RRR) The ratio of ARR to the event ppp=IEER-CERI/CER For very infrequent events, RR rate among controls can be large while AR is small Number needed to treat (NNT) Number of patients needed to NNT= 1 /ARR A good estimate of the effect receive a treatment for one size in easy-to-understand additional patient to benefit terms for patients

Number needed to treat (NNT) Number of patients needed to NNT= 1 /ARR A good estimate of the effect receive a treatment for one size in easy-to-understand additional patient to benefit terms for patients t Number needed to harm Number of patients needed to NNH= 1 /ARl ARI is the absolute risk increase (NNH) receive a treatment for one and equals I EER - CER I when additional patientto be the event is an unfavorable harmed outcome (e.g., drug side effect) An odds ratio (OR) estimates the odds of having or not having a particular outcome. When comparing therapeutic outcomes, in most cases OR can be substituted for RR as an equivalent measurement. AR, RR, and OR are estimates of the cumulative risk over time, usually defined at the end of the study period' DOil'T BE TR.ICKED . A disadvantage of RR is the potential for exaggeration. For example, interventions that reduce the rate of a disease from 4O'l. to 2O'/, and 4% to 27, each have an RR reduction of 5O%, but the ARR for the first case is 2o7' and the ARR for (Uo'2)'whereas the second case is only 2%. Based on the ARR, the NNT can be calculated' In the first case, the NNT is 5 the NNT for the second case is S0 (1/0.02).

DOil'T BE TR.ICKED . A disadvantage of RR is the potential for exaggeration. For example, interventions that reduce the rate of a disease from 4O'l. to 2O'/, and 4% to 27, each have an RR reduction of 5O%, but the ARR for the first case is 2o7' and the ARR for (Uo'2)'whereas the second case is only 2%. Based on the ARR, the NNT can be calculated' In the first case, the NNT is 5 the NNT for the second case is S0 (1/0.02). Confidence lnterval The CI provides boundaries within which exists a high probability (95'/o by convention) of finding the "true" value' For example' true value lies if the measured mean dilference between fvvo groups is 2.4 and the 95'l' CI is 1'9 to 3'0, the probability that the between 1.9 and 3.0 is 95'l.. for the control and When used in association with RR, if the CI includes the number 1, no risk or benefit exists; the outcomes experimental groups are statistically the same. 115