Practice-Based Learning and Improvement in Osteopathic Medical Practice
Practice-Based Learning and Improvement tests whether you can appraise evidence, apply it to one patient, and improve the system you work in. Level 3 items ask for the study design that answers a question, the statistic that describes benefit, or the quality-improvement step that comes next.
Study designs and levels of evidence
Meta-analysis of randomized trials sits above a single randomized trial, then cohort, case-control, cross-sectional, case series, and expert opinion. Randomization controls confounding, and intention-to-treat analysis preserves prognosis. Cohort studies yield incidence and relative risk but are vulnerable to confounding and loss to follow-up. Case-control studies suit rare diseases and are vulnerable to recall bias.
Test characteristics
Sensitivity is true positives divided by all diseased; specificity is true negatives divided by all healthy. Positive and negative predictive values depend on prevalence; sensitivity and specificity do not. The positive likelihood ratio is sensitivity divided by (1 minus specificity), and the negative likelihood ratio is (1 minus sensitivity) divided by specificity.
Measures of effect
Absolute risk reduction is the control event rate minus the treatment event rate; the number needed to treat is its reciprocal. Relative risk compares risk in exposed versus unexposed; the odds ratio is used in case-control designs; hazard ratios account for time to event.
Worked example
A screening test has 90% sensitivity and 95% specificity in 10,000 people with 2% prevalence. Of 200 diseased patients, 180 are true positives and 20 false negatives. Of 9,800 healthy people, 490 are false positives and 9,310 true negatives. Positive predictive value is 180/670, or 27%; negative predictive value is 9,310/9,330, or 99.8%. If prevalence rises to 20%, positive predictive value rises to about 82% with identical sensitivity and specificity.
Quality improvement
Use the Model for Improvement: three questions (what are we trying to accomplish, how will we know change is an improvement, what changes can we make) plus rapid PDSA cycles. Run charts and control charts separate common-cause from special-cause variation.
Traps
Equating positive predictive value with sensitivity. Concluding that p above 0.05 proves no effect, or that a surrogate endpoint equals clinical benefit. Calling a before-and-after audit a randomized trial.