Reviewers' Comments

Reason for Consult: Questions About Case-Control Design

Manuscript: “Relationship between Reasons for Disciplinary Action by State Licensing Boards and Types of Unprofessional Behaviors”

Editor's Comments: The study employed a case-control design to study the association between reasons for disciplinary action by state licensing boards and types of unprofessional behaviors. The article was well written, though a couple of analytical issues should have been described with more details or clarity.

• First, in this study, each case is matched with two controls. Although Table 1 offers an overall summary of the data, a more informative univariate analysis can be presented by appropriate tests that account for the 1:2 matching structure. In addition, one or two more sentences should be added to describe the rationale of conditional logistic regression since it is not as popular as the unconditional one. I suggest a separate section be created to discuss statistical methods used in the analysis under the Methods section.
  
• Second, is there any interaction between the types of unprofessional behaviors and other covariates being adjusted (e.g. sex)? A more thorough analysis by including appropriate interaction terms and/or stratification might be helpful to enhance the understanding of the association being studied.
  
• Third, the appropriate interpretation of the results, which is also related to the limitation of the case-control design, should be more clearly conveyed. Although several odds ratio suggest an association between unprofessional behaviors and disciplinary action, it should be emphasized that the significance or impact of unprofessional behaviors at medical school on future working behaviors at the population level also depends on factors that could not be captured by a case-control design such as absolute risk and prevalence of exposure. For example, if the absolutely risk of disciplinary action for subject with/without unprofessional behaviors is 3% and 1%, respectively, then the odds ratio is around 3, suggesting an association. Nevertheless, the absolute risk is still only 3%. If the prevalence of the unprofessional behavior is low, then the problem becomes even more minor.

Consultant's Advice:

This consult was scheduled to give advice on statistical aspects of a recent paper submission. The paper has requested.

1. A justification for the conditional logistic regression approach. I have furnished 3 references which explain that this is the desirable analysis for matched case control studies.

http://ije.oxfordjournals.org/cgi/content/abstract/9/1/89 - Pike MC, Hill AP, Smith PG. Bias and efficiency in logistic analyses of stratified case-control studies. International journal of epidemiology 1980;9(1):89-95.

http://epirev.oxfordjournals.org/cgi/reprint/16/1/33 - Thompson WD. Statistical analysis of case-control studies. Epidemiologic reviews 1994;16(1):33-50.

http://tinyurl.com/yum2pd - Breslow N. Design and analysis of case-control studies. Annual review of public health 1982;3:29-54.

2. P-values in table 1. The appropriate OR and test should be the OR for matched studies and McNermar’s test. This is covered on p. 40 of the Thompson paper. A simple way to obtain this would be to perform a conditional logistic regression with each of the predictors in Table 1 added one-at-a-time as a single factor. This would be the simplest approach since SPSS has very limited capabilities.

3. Interactions: The investigator was advised to consider which would be plausible and then return for assistance in fitting and interpreting them.

4. Practical significance: The investigator was advised to consider this issue seriously. If she can obtain an estimate of the % of graduates who are disciplined, then it is possible to obtain a proportion of people with the risk factor who will be disciplined. This is helpful addition.