Dataset Information

A Bayesian latent class approach for EHR-based phenotyping.

ABSTRACT: Phenotyping, ie, identification of patients possessing a characteristic of interest, is a fundamental task for research conducted using electronic health records. However, challenges to this task include imperfect sensitivity and specificity of clinical codes and inconsistent availability of more detailed data such as laboratory test results. Despite these challenges, most existing electronic health records-derived phenotypes are rule-based, consisting of a series of Boolean arguments informed by expert knowledge of the disease of interest and its coding. The objective of this paper is to introduce a Bayesian latent phenotyping approach that accounts for imperfect data elements and missing not at random missingness patterns that can be used when no gold-standard data are available. We conducted simulation studies to compare alternative phenotyping methods under different patterns of missingness and applied these approaches to a cohort of 68 265 children at elevated risk for type 2 diabetes mellitus (T2DM). In simulation studies, the latent class approach had similar sensitivity to a rule-based approach (95.9% vs 91.9%) while substantially improving specificity (99.7% vs 90.8%). In the PEDSnet cohort, we found that biomarkers and clinical codes were strongly associated with latent T2DM status. The latent T2DM class was also strongly predictive of missingness in biomarkers. Glucose was missing in 83.4% of patients (odds ratio for latent T2DM status = 0.52) while hemoglobin A1c was missing in 91.2% (odds ratio for latent T2DM status = 0.03 ), suggesting missing not at random missingness. The latent phenotype approach may substantially improve on rule-based phenotyping.

SUBMITTER: Hubbard RA

PROVIDER: S-EPMC6519239 | biostudies-literature |

REPOSITORIES: biostudies-literature

ACCESS DATA

Similar Datasets

Project description:AimsThe main focus of this study is to illustrate the importance of the statistical analysis in the evaluation of the accuracy of malaria diagnostic tests, without admitting a reference test, exploring a dataset (n=3317) collected in São Tomé and Príncipe.MethodsBayesian Latent Class Models (without and with constraints) are used to estimate the malaria infection prevalence, together with sensitivities, specificities, and predictive values of three diagnostic tests (RDT, Microscopy and PCR), in four subpopulations simultaneously based on a stratified analysis by age groups (< 5, ? 5 years old) and fever status (febrile, afebrile).ResultsIn the afebrile individuals with at least five years old, the posterior mean of the malaria infection prevalence is 3.2% with a highest posterior density interval of [2.3-4.1]. The other three subpopulations (febrile ? 5 years, afebrile or febrile children less than 5 years) present a higher prevalence around 10.3% [8.8-11.7]. In afebrile children under-five years old, the sensitivity of microscopy is 50.5% [37.7-63.2]. In children under-five, the estimated sensitivities/specificities of RDT are 95.4% [90.3-99.5]/93.8% [91.6-96.0]--afebrile--and 94.1% [87.5-99.4]/97.5% [95.5-99.3]--febrile. In individuals with at least five years old are 96.0% [91.5-99.7]/98.7% [98.1-99.2]--afebrile--and 97.9% [95.3-99.8]/97.7% [96.6-98.6]--febrile. The PCR yields the most reliable results in four subpopulations.ConclusionsThe utility of this RDT in the field seems to be relevant. However, in all subpopulations, data provide enough evidence to suggest caution with the positive predictive values of the RDT. Microscopy has poor sensitivity compared to the other tests, particularly, in the afebrile children less than 5 years. This type of findings reveals the danger of statistical analysis based on microscopy as a reference test. Bayesian Latent Class Models provide a powerful tool to evaluate malaria diagnostic tests, taking into account different groups of interest.

Project description:Rift Valley Fever is an important zoonotic viral disease of livestock occurring across much of Africa causing acute febrile illness, abortion, and neonatal death in livestock particularly sheep and cattle and a range of disease in humans from mild flu-like symptoms to more severe haemorrhagic fever and death. Understanding the epidemiology requires well-evaluated tools including antibody detection ELISAs. It is well-recognized that tests developed in one population do not necessarily perform as well when used in different populations and it is therefore important to assess tests in the populations in which they are to be used. Here we describe the performance of a commercial RVF ELISA (ID.Vet) and an in-house plaque reduction neutralization test (PRNT80). A Bayesian no gold standard latent class model for two tests and ?2 populations based on the Hui-Walter model was used to estimate the test parameters using a range of populations based on geographical separation and age to assess consistency of performance across different sub-populations. The ID.Vet ELISA had an estimated diagnostic sensitivity (Se) of 0.854 (0.655-0.991 95%BCI) and specificity (Sp) of 0.986 (0.971-0.998 95%BCI) using all the data and splitting the population by geographical region compared to 0.844 (0.660-0.973 95%BCI) and 0.981 (0.965-0.996 95%BCI) for the PRNT80. There was slight variation in the mean Se and Sp in different sub-populations mainly in Se estimates due to small numbers of positives in the sub-populations but the 95% BCI generally overlapped suggesting a very consistent performance across the different geographical areas and ages of animals. This is one of few reports of serological evidence of RVF in Central Africa and strongly suggests the virus is actively circulating in this cattle population. This has important public health implications and RVF should be considered as a differential in both livestock disease cases as well as human febrile cases in West and Central Africa not just East Africa. We also demonstrate that the performance of the commercial ELISA is comparable to the PRNT80 but has the advantages of speed, lower cost and no containment needs making it a much more useful test for low and middle income settings (LMICs).

Dataset Information

A Bayesian latent class approach for EHR-based phenotyping.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets