Estimating the burden of underdiagnosis within England: A modelling study of linked primary care data

Undiagnosed chronic disease has serious health consequences, and variation in rates of underdiagnosis between populations can contribute to health inequalities. We aimed to estimate the level of undiagnosed disease of 11 common conditions and its variation across sociodemographic characteristics and regions in England.

We used linked primary care, hospital and mortality data on approximately 1.3 million patients registered at a GP practice for more than one year from 01/04/2008-31/03/2020 from Clinical Practice Research Datalink. We created a dynamic state model with six states based on the diagnosis and mortality of 11 conditions: coronary heart disease (CHD), stroke, hypertension, chronic obstructive pulmonary disease, type 2 diabetes, dementia, breast cancer, prostate cancer, lung cancer, colorectal cancer, and depression/anxiety. Undiagnosed disease was conceptualised as those who died with a condition but were not previously diagnosed. This was combined with observed data on the incidence of diagnosis, the case fatality rate in the diagnosed, and an assumption about how that rate varies with diagnosis to estimate the number of undiagnosed disease cases over the total number of disease cases (underdiagnosis) in each population group. We estimated underdiagnosis by year, sex, 10-year age group, relative deprivation, and administrative region. We then applied small-area estimation techniques to derive underdiagnosis estimates for health planning areas (CCGs).

Levels of underdiagnosis varied between 16% for stroke and 69% for prostate cancer in 2018. For all diseases, the level of underdiagnosis declined over time. Underdiagnosis was not consistently concentrated in areas with high deprivation. For depression/anxiety and stroke, underdiagnosis was estimated to be higher in less deprived CCGs, whilst for CHD and T2DM, it was estimated to be higher in more deprived CCGs, with no apparent relationships for other conditions. We found no uniform spatial patterns of underdiagnosis across all diseases, and the relationship between age, deprivation and the probability of being undiagnosed varied greatly between diseases.

Our findings suggest that underdiagnosis is not consistently concentrated in areas with high deprivation, nor is there a uniform spatial underdiagnosis pattern across diseases. This novel method for estimating the burden of underdiagnosis within England depends on the quality of routinely collected data, but it suggests that more research is needed to understand the key drivers of underdiagnosis.