Harnessing electronic health records to monitor the progression of chronic diseases

AE1: Patient records and electronic health records AE2: Chronic disease progression including CKD, CPD AE3: Mathematical models AE4: Statistical analysis

EHR data

Dementia: D-CRIS contains data on N~7,000 with at least 3 MMSE observations, current medication use will be used as a proxy for co-morbidities.

Chronic Obstructive Pulmonary Disorder (COPD): CPRD contains data on N ~ 117k COPD patients with at least two FEV1 observations, plus co-morbid diagnoses.

Chronic Kidney Disease (CKD): SIR (primary and secondary) contains creatinine and co-morbidity data on over 23k patients with CKD.

Objectives:

1. Model disease progression and the effect of co-morbidities using standard approaches stratified by disease stage or type
2. Investigate the utility of alignment approaches for characterizing the effect of co-morbidities on disease progression
3. Examine whether latent classes can be used to identify clinically meaningful subgroups with different progression characteristics
4. Use findings to extend existing disease specific models and develop personalised medicine approaches

Methods
Objective 1: Mixed effect models of patients with same disease stage or type at first visit. Joint longitudinal and survival models will be used to assess impact of missing data due to death.

Objective 2: Villemagne alignment will be tested in a simulation study, before being applied to study patients who have the relevant diagnosis by last visit. Regression will be used to relate model residuals to co-morbidities.

Objective 3: The existence of subgroups will be assessed by looking for multimodality in distributions from objectives 1 & 2. If evidence for subgroups exist, latent class extensions to alignment or mixed effect models will be used.

Objective 4: Models from objectives 1 - 3 will be used to extend established disease progression models, which will then be used for decision support tools.

Potential additional objectives: Study individual-level treatment effect in EHR embedded trials or in observational EHR treatment studies. Additional use cases, e.g. visual acuity in macular degeneration.

Who will benefit from this research?

This research has the potential to affect the lives of the 1.8m individuals with Chronic Kidney Disease (CKD), the 500,000 individuals with Chronic Obstructive Pulmonary Disorder (COPD) and the 850,000 individuals with dementia in the UK. It also has the potential to benefit the UK healthcare system, by reducing costs due to suboptimal care for patients, e.g. decision support tools based on our findings could streamline and complement clinical decisions in relation to care for patient with co-morbidities. This is demonstrated by the commitment by five clinical academics (Iain Buchan, Jennifer Quint, Rob Stewart, John O'Brien and Rob Howard) to collaborate, as per their letters of support.

By highlighting specific co-morbidities which have a differential effect on disease progression, we will help the pharmaceutical industry to design trials and identify individuals who could most benefit from novel treatment approaches. This is highlighted by Lindsay Edward's letter of support; Prof Edward is head of respiratory data science for GSK, whose commitment to EHR research on COPD is demonstrated by their EHR embedded Salford Lung Study. Eventually, accurate disease specific progression models may make such health record embedded clinical trials for chronic diseases more effective, with the potential to reduce costs for both academic and pharmaceutical clinical trial researchers.

Potential outcomes from this research:

(1) Guidelines for modelling chronic disease progression using EHRs
(2) Prognostic models for co-morbid patients, that aid in the design of clinical trials
(3) Decision support tools complementing clinical decision making in relation to co-morbidities
(4) A multidisciplinary group cross-fertilising methods and ideas between UK investments in methodology research (e.g. MRC Biostatistics Unit), dementia research (e.g. DPUK) and e-Health research (Farr institute)

How will they benefit from this research?

Researchers (academic and industry): In the short-term clinical epidemiologists studying chronic diseases will benefit from modelling guidelines (objectives 1 - 3), while biostatisticians will be able to extend methods we study or test their own methodology on EHR data that we highlight and characterise. Methodological guidelines I produce will also help researchers using conventional cohorts to study the progression of chronic disease, e.g. in dementia prevention studies such as the EPAD register [20] or Brain Health Registry (http://www.brainhealthregistry.org/). In the medium-term health informaticians will be able to use the models from objective 4 to inform clinical decision support tools to complement clinical descision making, while clinical trial designers will be able to take advantage of them to increase the utility of trial simulations. Notably, these trial simulations will allow the effect of inclusion or exclusion of co-morbid patients to be studied, allowing trial design to improve. Researchers in biomarker technology companies will be better able to highlight the potential utility of their tests for secondary EHR research.

Clinicians: In the short-term clinicians will be better informed about average progression and the effect of co-morbidities on progression (objectives 1 - 2). In the medium term these models could help to inform their clinical practice through decision support tools (objectives 3 - 4). In the long term, these models could enable EHR embedded clinical trials, moving the NHS closer to a Learning Health System (LHS) model where every patients data is used to improved future care (http://www.learninghealthcareproject.org/).

Patients: will benefit from clinicians who are more informed about the effect of co-morbidities on prognosis, allowing them to provide more patient-centred care. In the longer term, they could benefit from tailored treatments identified and demonstrated within a LHS framework.