User Tools

Site Tools


observ

Observational studies

Confounding adjustment (ANCOVA)

The development of the statistical model also based on p-values, “Variables that reached a p-value <.1 in the univariate analyses were entered into a multivariate model” and “the best model was chosen based on its predictive power and Akaike Information Criterion”. However, the estimation of risk factor effects must be based on considerations of cause-effect relationships, not on statistical precision (p-values) because while the inclusion of a confounder in a statistical model is likely to reduce bias, the inclusion of a mediator or collider can induce bias, see Shrier I, Platt RW. Reducing bias through directed acyclic graphs. BMC Med Res Methodol 2008;8:70. Statistical precision may have a role to play when developing prediction models, but they are not useful for developing models for estimation of risk factors. Please provide a rationale for the selection of covariates in terms of cause and effect.

Confounding adjustment (PS)

The propensity scores have been calculated using a statistical model with several variables. Please describe how these variables are measured and provide a rationale for their selection. Please clarify also if any of the variables could induce adjustment bias, see e.g. Sauer B, Brookhart MA, Roy JA, VanderWeele TJ. Developing a Protocol for Observational Comparative Effectiveness Research: A User’s Guide. Rockville (MD), Agency for Healthcare Research and Quality (US); 2013 Jan. Publication No.: 12(13)-EHC099. Chapter 7. Covariate Selection.

The Table 2 fallacy

Table 2 seems to reflect a case of the so-called Table 2 fallacy, see Westreich D, Greenland S. The table 2 fallacy: presenting and interpreting confounder and modifier coefficients. Am J Epidemiol 2013;177:292-298. Please describe the fitted models more clearly.

Primary endpoints

A primary outcome measure is presented. Structuring of endpoints into primary and secondary is usually part of a strategy for addressing multiplicity issues in randomised trials but is hardly relevant in an observational study as the one described here. Please clarify the terminology.

Stepwise regression

Statistical models are developed using stepwise regression, a method having precision (p-values) as selection criteria. It ignores cause-effect relations. However, while the inclusion of a confounder into a statistical model may reduce bias, the inclusion of a mediator or collider induces adjustent bias. If the purpose of the model is to provide clinically interpretable parameter estimates, stepwise regression is an inadequate method.

observ.txt · Last modified: 2020/04/13 09:09 by ranstam