Exploring the use of ‘landmarking’ methods to explain the variation of COPD mortality risk over time

Start Date: 1 Aug 2015 End Date: 1 Oct 2016

Exploring the use of ‘landmarking’ methods to explain the variation of COPD mortality risk over time

The problem

Chronic Obstructive Pulmonary Disease (COPD) is a disabling condition with high prevalence from middle age onwards. The DOSE index combines clinical features of COPD (dyspnoea, obstruction, smoking and exacerbation's) to produce an overall measure of disease severity.

Dyspnoea is measured using the MRC breathlessness score; obstruction by FEV1 (forced expiratory volume in 1 second); smoking status and exacerbation history by the occurrence of disease worsening events within the previous year. Values of DOSE index can range between 0 and 8. Landmarking methods can produce dynamic predictions for a given event occurring within a given time horizon.

The solution

We used data from the Hampshire Health Record Analytical database (HHRA), to investigate whether all-cause mortality risk varies according to levels of the DOSE index. The data contained anonymised primary and secondary care data for around 1.4 million patients living in Hampshire, UK, selecting all prevalent COPD patients at 1st January 2010 with at least one recorded DOSE index score (N=10017), following them up for five years. 


We found that COPD patients having a DOSE index of 4 or 5 (relative to 3 or less) were estimated to have between 2 and 6 times the risk of dying, with the relative hazard reducing with age. Patients having a DOSE index of 6 to 8 generally showed a still higher mortality risk, but estimation was imprecise.  When assessing risk for DOSE index scores at lower levels (less severe disease), we found that even a single extra DOSE index point was associated with a significant mortality risk in patients aged in their mid-70s. Having a DOSE index of 2 or 3 points showed a statistically significant increase in mortality risk at almost all ages against those with a DOSE index of zero.

This work was presented as a poster at the 2016 annual conference of the International Society for Clinical Biostatistics (ISCB). 

Dr David Culliford