Screening models

Predicting age of onset for chronic diseases

Healthcare, Medical Research

An independent organisation in primary care and prevention realised that under increased scrutiny, the efficacy of screening programmes appears questionable.

Download PDF

Goal

Healthcare costs represent the largest category of GDP and are continuing to rise. The early detection of chronic diseases—conditions, such as diabetes or hypertension—that require sometimes costly and sustained treatment have been proven to slow the growth of healthcare costs. An independent organisation in primary care and prevention wished to better predict the age of onset for chronic diseases, reduce unnecessary patient interventions,and slow the rise of healthcare costs.

Insight and Action

The organisation brought in QuantumBlack to look through their data sets and improve efficiency. Fortunately, rich, fine-grained data sets that may enable more targeted policies are increasingly available. Personalised, data-driven screening recommendations offer a potential mechanism for improving the efficiency of healthcare delivery.

We developed age-of-onset models for a number of chronic conditions, including hypertension, prostate cancer, breast cancer, and stroke, using margin-based, censored-regression ensembles. The models incorporated about 900 covariates—mostly patient biomarkers and socio-economic factors—on about 20,000 patients tracked longitudinally over nearly 15 years.

Results

  • 99Percentof hypertension and stroke instances were detected by a simple screening using the model.
  • 11Percentreduction in unnecessary interventions for patients 65 years and older through the prostate age-of-onset model’s screening filter.
  • 2.7Yearsrepresents the average error achieved for hypertension and stroke models for the age of onset.