論文を批判的に読むときのチェックリスト

主に医学・疫学・社会科学系の論文を読むときに、こんなことに注目しながら査読やknowledge gap探し（先行研究の弱点を見つけること）をしていますというリストです

Transcript

1. 1. Question:
   • Is the research novel?
   • Is it important to public health and addressing health disparities?
   • Do the study’s methods provide an answer to this?
   • Watch out for any disconnection between methods and questions: this is
   surprisingly common.
   2. Sampling:
   • Consider who’s included and who’s excluded (i.e., "who’s on the map and who’s
   off the map").
   • Consider the implications on the target population and health disparities.
   • Look for selection bias.
   3. Measurement:
   • Assess the concordance between the research question/hypothesis and what’s
   measured.
   • Check the validity and accuracy of measurements.
   4. Causal Estimand:
   • Is the study estimating ATE (Average Treatment Effect), CATE (Conditional
   Average Treatment Effect), LATE (Local Average Treatment Effect) or else?
   • What is the hypothetical intervention being considered?
   • Cumulative treatment vs incident treatment
   • Time-fixed vs time-varying
   • Are these choices explicit? Does the Discussion acknowledge the choice?
   • Any disconnection between the estimand and the discussion/interpretation?
   5. Identification:
   • What are the assumptions? Are they plausible and explicitly acknowledged?
   • Confounding
   • measurement errors and residual confounding
   • unobserved confounders
   1. Is the bias big conditional on adjusted covariates?
   • Selection bias
   • Attrition (before or after the exposure?)
   • How is missing data handled?
   • (Differential) measurement errors
   • Check for ambiguous “time zero”.
   6. Estimation:
   • Watch for potential misspecification, such as non-linearity.
   7. Interpretation:
   • Look for evidence of confounding in Table 1.
   • Look for evidence of selection bias.
   • Watch for authors misinterpreting their results.
   • Treating the absence of evidence as evidence of absence.
   • Making a false dichotomy of p-values.
   • Generalizing results too much without considering causal estimand and
   potential effect heterogeneity.
   • Consider potential mechanisms/policy implications.
   • Consider implications on health disparities.
   

   View Slide