A novel case‐control subsampling approach for rapid model exploration of large clustered binary data

Abstract

In many settings, an analysis goal is the identification of a factor, or set of factors associated with an event or outcome. Often, these associations are then used for inference and prediction. Unfortunately, in the big data era, the model building and exploration phases of analysis can be time-consuming, especially if constrained by computing power (ie, a typical corporate workstation). To speed up this model development, we propose a novel subsampling scheme to enable rapid model exploration of clustered binary data using flexible yet complex model set-ups (GLMMs with additive smoothing splines). By reframing the binary response prospective cohort study into a case-control-type design, and using our knowledge of sampling fractions, we show one can approximate the model estimates as would be calculated from a full cohort analysis. This idea is extended to derive cluster-specific sampling fractions and thereby incorporate cluster variation into an analysis. Importantly, we demonstrate that previously computationally prohibitive analyses can be conducted in a timely manner on a typical workstation. The approach is applied to analysing risk factors associated with adverse reactions relating to blood donation.

Outcome-Dependent Sampling in Cluster-Correlated Data Settings with Application to Hospital Profiling

“…patients for whom Y ki =0) and n 1 k cases (i.e. patients for whom Y ki =1) from the N 0 k non‐cases and N 1 k cases respectively in the k th hospital (Wright et al ., ; Haneuse and Rivera‐Rodriguez, ). To realize the gains of an ODS design, we must resolve the fact that the individuals who have ‘complete’ data are no longer representative of the underlying population.…”

“…As indicated in Section 1, Wright et al . () also considered estimation and inference for a GLMM based on data collected via a CSCC sampling scheme. Briefly, let ξ k = .…”

“…Wright et al . () further considered this design in ‘big data’ settings when interest is in an LNGLMM, but actually fitting the model to the entire data set is computationally infeasible. They proposed that inference proceed by including cluster‐specific offset terms in the regression model, although the derivation of the offset terms hinges on the choice of the logit link function.…”

“…Second, we find that the offsetted regression approach of Wright et al . () significantly underestimates between‐hospital variability and hence is unsuitable for profiling. We thus adopt the PML approach and extend it to compute hospital‐specific quality measures.…”

Statistical strategies for the analysis of massive data sets

“…What was particularly interesting in the Wright et al. () analysis was that they did a clustered analysis, treating donor center, as the cluster. With a very simple adjustment via inclusion of an offset, models could be easily fit using R functions glm or gam using only a desktop computer.…”