Characterisation of general practices by prescribing and diagnostic test ordering behaviour using dimensionality reduction and clustering analysis

Introduction

Personality refers to an individual’s enduring characteristics and behaviours. Personality characteristics can be inferred from behaviour and it has been demonstrated that Facebook users’ likes can be used to reliably identify their personality characteristics and preferences.[i] In an analogous way, general practitioners (GPs) and general practices may have enduring, characteristic ways of practicing medicine, which can be identified from their clinical behaviour.

Research in the Netherlands in the 1990s characterised GPs by their patterns of prescribing, use of diagnostic tests and referral rates.[ii] These were further investigated through observation of their consultation style. Three distinct typologies or GP practice styles were described: GPs with an integrated consultation style, with low diagnostic test, prescription and referral rates; an interventionist style with high diagnostic test, prescription and referral rates; and a minimal diagnostic style, characterised by low diagnostic test rates, but high prescription and referral rates. These practice styles were associated with health outcomes. Patients of GPs with an integrated style consulted less frequently and self-reported better health. The research was undertaken on a small sample of GPs and patients.

More recently, analyses of large datasets of prescribing data have observed strong correlations of prescribing rates of quite different types of drugs by primary care prescribers. In a US study, benzodiazepine prescribing was highly correlated with opioid prescribing, antibiotic prescribing and prescribing of high-risk medications for the elderly.[iii] This suggests a single underlying construct (propensity to prescribe) across a range of drugs. The authors suggested this might be due to an underlying physician characteristic or the physician’s characteristic consultation style.

A US analysis identified four clearly differentiated groups of primary care physicians, through analysis of their prescribing rates of benzodiazepines, opioids, proton pump inhibitors and antibiotics [iv]

A similar UK study analysed two databases, one of primary care records, the other of prescription reimbursement data. It also found strong correlations at the level of general practice between antibiotic prescribing rates and prescribing rates of all other drugs.[v] These correlations were only modestly attenuated after adjustment for patient characteristics. The authors concluded that propensity to prescribe was the single strongest determinant of antibiotic prescribing.

Analysis of diagnostic test rates in primary care has also shown strong correlations between rates of ordering of some apparently unrelated diagnostic blood tests (full blood counts, renal function tests, liver function tests).[vi] This also suggests an underlying propensity to undertake certain diagnostic tests.

Research has investigated the role of physician cognitive characteristics and specific propensities to prescribe and to request diagnostic tests.[vii] [viii],[ix] More widely, it has been observed that investigation of variation in health care use has overlooked the role of style of physician practice.[x]

There is evidence that GP prescribing rates and diagnostic test ordering rates are influenced by underlying propensities to prescribe and order diagnostic tests. These characteristics may also be determinants of health care use and of patient outcomes.

Aim

The doctoral thesis aims to will investigate whether clinical prescribing and diagnostic test-ordering behaviour can be used to characterise GPs and general practices and to predict health care outcomes.

Objectives

1.      Investigate patterns of drug prescribing and diagnostic-test ordering across individual GPs using dimensionality reduction methods.

2.      Investigate patterns of drug prescribing and diagnostic-test ordering across general practices using dimensionality reduction methods.

3.      Investigate the characterisation of GPs and general practices using prescribing and diagnostic test ordering behaviour using clustering analysis.

4.      Investigate whether drug prescribing and diagnostic-test ordering characteristics of general practices predict health outcomes.

Method

This project will analyse a large dataset of electronic health records from UK primary health care to investigate whether GPs and general practices have underlying propensities to prescribe and to request diagnostic tests; whether GPs’ and general practices’ prescribing and diagnostic test ordering behaviour are enduring characteristics; and whether these characteristics can be use to describe GPs and general practices.

Skills required

The project requires a student with strong quantitative skills in data analysis and handling of large datasets.