Abstract
The present study aimed to compare the predictive acuity of latent class regression (LCR) modelling with: standard generalised linear modelling (GLM); and GLMs that include the membership of subgroups/classes (identified through prior latent class analysis; LCA) as alternative or additional candidate predictors. Using real world demographic and clinical data from 1,802 heart failure patients enrolled in the UK-HEART2 cohort, the study found that univariable GLMs using LCA-generated subgroup/class membership as the sole candidate predictor of survival were inferior to standard multivariable GLMs using the same four covariates as those used in the LCA. The inclusion of the LCA subgroup/class membership together with these four covariates as candidate predictors in a multivariable GLM showed no improvement in predictive acuity. In contrast, LCR modelling resulted in a 18–22% improvement in predictive acuity and provided a range of alternative models from which it would be possible to balance predictive acuity against entropy to select models that were optimally suited to improve the efficient allocation of clinical resources to address the differential risk of the outcome (in this instance, survival). These findings provide proof-of-principle that LCR modelling can improve the predictive acuity of GLMs and enhance the clinical utility of their predictions. These improvements warrant further attention and exploration, including the use of alternative techniques (including machine learning algorithms) that are also capable of generating latent class structure while determining outcome predictions, particularly for use with large and routinely collected clinical datasets, and with binary, count and continuous variables.
Highlights
The limited acuity and clinical utility of generalised linear models (GLMs)The potential utility of predictive modelling, using routinely collected data for diagnosis, prognostication and health service planning, is one of five ‘novel capabilities’ that Wang et al [1] identified as pertinent to the application of data analytics in medicine and health
While clinical prediction models (CPMs) and their wider utility remain contentious, many of the standard statistical modelling techniques commonly used are on clinical datasets that remain relatively small—at least when compared to contemporary notions of ‘Big Data’ [2]
In Procedure 3, the second step involved consideration of the four covariates as candidate predictors of survival in the Cox proportional hazards model, and membership of the same 5-class model developed in the first step of Procedure 2. These analyses found that: the best fitting GLMs did not retain class membership as a predictor; and forcibly retaining class membership in the model did not improve the c-statistics compared to what was achieved in Procedure 1, regardless of how class membership was assigned
Summary
The limited acuity and clinical utility of generalised linear models (GLMs)The potential utility of predictive modelling, using routinely collected data for diagnosis, prognostication and health service planning, is one of five ‘novel capabilities’ that Wang et al [1] identified as pertinent to the application of data analytics in medicine and health. The limited acuity and clinical utility of generalised linear models (GLMs). While CPMs and their wider utility remain contentious (beyond strict prognostication, and in prevention [6–9]), many of the standard statistical modelling techniques commonly used are on clinical datasets that remain relatively small—at least when compared to contemporary notions of ‘Big Data’ [2]. A substantial statistical weakness of the commonest of these (generalised linear models; GLMs) as a predictive tool is that they often fail to make full use of the joint information available amongst all candidate predictor variables. This is because these models rarely explore nonlinear relationships and interactions. Even when analysts optimally parameterise the candidate predictors available, and carefully consider all possible interaction terms between these, the clinical utility of GLMs is typically limited to predictions made at the population level [6,10], while predictions at the individual level often lack precision (and with it, utility)
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have