Table 1

Variability in CPRD GOLD and HES and their potential causes and solutions

FindingObservable causePossible original causePossible solutions
CPRD GOLD
Gradual change in the population distribution between 2001 and 2007Increases in the prevalence of recorded cardiovascular diseaseDemographic changes (eg, ageing); incremental improvements in diagnostic coding or in clinical proceduresIncremental learning of models; inclusion of time interaction effect
Shift in the direction of change in 2008After the previous year’s increase, the prevalence of CHD, heart failure and PAD started decreasing again around the same timeNo immediate reason identifiedSeparate analyses of prechange and postchange data
Oscillations in the data distributions after 2010Changes in the distribution of socioeconomic status in the target distributionSelective dropout of practices, possibly related to a switch in the practice management softwareMixed models with practice effects
HES
Gradual change in the population distribution between 2001 and 2008Increase in reported chronic CHD and atrial fibrillation; decreases in reported angina pectoris, acute myocardial infarction, heart failure and strokeDemographic changes (eg, ageing); incremental improvements in diagnostic coding or in clinical procedures; selective increase of disease incidenceIncremental learning of models; inclusion of continuous time interaction effect
Shift in the direction of change in 2009Increased coding of transient cerebral ischaemic attacks between 2009 and 2010No immediate reason identifiedSeparate analyses of prechange and postchange data
Abrupt change in March 2010Drop in acute myocardial infarction codingUpdate to the HSCIC Coding Clinic Guidance in February 2010Separate analyses; incremental learning of models
Further abrupt changes in April 2012 and 2014Sudden increase in acute myocardial infarction coding in 2012 with concomitant drop in subsequent myocardial infarction records; sudden further decrease in angina pectoris codes in 2014Update to the National Clinical Coding Guidance National Clinical Coding Standards ICD-10 fourth EditionSeparate analyses; incremental learning of models
  • CHD, coronary heart disease; CPRD GOLD, Clinical Practice Research Datalink; HES, Hospital Episode Statistics; ICD-10, International Classification of Diseases 10th revision; PAD, peripheral arterial disease.