Collecting morbidity data in general practice: the Somerset morbidity project

Introduction

An important responsibility of health authorities and general practitioners is to assess the health needs of their local populations and to commission services to meet these needs. This task is hampered by a lack of routinely available, up to date, and accurate information on incidence and prevalence of disease. Existing national sources of morbidity data, of which the most comprehensive are the national morbidity studies,1 2 3 4 are not representative of smaller localities or health authority populations, may be out of date as the studies take place only decennially, and may be biased by the fact that only general practitioners who have volunteered contribute.

General practice records have already been recognised as a potentially rich source of morbidity data.5 Improved uptake of computerisation in primary care and technological advances in electronic data transfer make it possible to use general practice computer systems for assessing local health needs. However, attempts at collecting up to date and reliable information in other areas of Britain have had only limited success because of problems with completeness and accuracy of the data, Read codes, computer software, and information technology, particularly arising as a result of the wide variety of commercially available general practice computer systems.6 7 8

The Somerset morbidity project aims to collect valid, complete, continuous, and representative data on disease presenting to primary care. The resulting database is readily obtainable; useful to purchasers, providers, and general practitioners; and could be reproduced in other health districts. We report on the data collection methods and analyse the data collected on three conditions in greater detail: glue ear, where the health authority has used the data to commission surgical services; asthma, which has generated interest among the general practitioner user group; and hypertension, where the local prevalence seemed to be significantly different from the prevalence found nationally.

Methods

To avoid the potential bias of using volunteer general practitioners we selected practices with a suitable computer system that were representative of the district health authority population for general practitioner, practice, and population characteristics. In 1993, when recruitment began, 97% of practices in Somerset were computerised and 64% were using a computer system suitable for our study (using Read codes and software compatible with our data extraction software). We recruited from these practices in four stages using a computerised model established for the purpose. The model initially generated all possible samples of practices that met the representativeness criteria within predefined acceptable limits. Those practices appearing most frequently in the samples were approached and recruited. As practices were recruited the sampling process was repeated.

We chose this method as it was systematic, repeatable by others, and maximised the chances of selecting a representative sample if a substantial number of practices declined to participate. However, only one practice we approached declined. We treated the resulting sample as a randomly selected sample of clusters of unequal size for the purpose of calculating confidence intervals for population characteristics using the method described by Kish.9

Each practice is paid a small fee, currently 40 pence a year for each patient registered with the practice, which covers the cost of generating the weekly report.

RECORDING AND EXTRACTING DATA

Practices record Read and episode type codes for each face to face contact between patients and practice medical and nursing staff. The episode types refer to whether the patient is consulting with the illness for the first time (F), attending for a new episode of a previously diagnosed illness (N), or for a follow up consultation (O).

The extraction software searches for each diagnosis for each patient during the time specified. It also prioritises episode types for each diagnosis: F over N, over O. Each diagnosis is therefore represented only once for a patient for the time concerned. Data are routinely extracted at the end of each week, the end of each quarter, and the end of each year. These searches produce different databases that enable disease processes with different epidemiological patterns to be studied optimally.

The data are sent electronically by modem to an independent collection agency (the Royal College of General Practitioners Birmingham Research Unit) and then on to the database held at the health authority. No identifiable patient details leave the practice, and the research unit gives data from each practice a code, which is known only to the project coordinator. Analyses are presented coded by practice, and each practice knows only its own code.

DATA QUALITY

Several systems are in place to ensure optimum data quality. Firstly, all general practitioners and practice staff are given initial training and regular updates. Secondly, the project coordinator visits each practice on a random day every three months to validate the data. The Read codes entered for a random 25 patient sample are checked against the paper notes to ensure that they are compatible, appropriate, and given correct codes and episode types. Thirdly, the computer carries out edit checks for rare diagnoses and those recorded outside the usual age and sex parameters—for example, gynaecological disorders in men, very rare diagnoses where a coding error is more likely than the disease—for example, tetanus immunisation wrongly coded as a case of clinical tetanus. Incorrectly coded items noted during the validation visits and as a result of the edit checks are brought to the practice's attention and the computerised records amended. Data integrity checks are also carried out to verify the electronic interchange process, and a general practice user group meets regularly to enable data feedback and interpractice comparisons and to agree common use of Read codes where appropriate across the participating practices.

Results

We recruited a final sample of 11 practices with a combined population of about 65 000. The practices were representative of the health authority for geographical spread, urban-rural balance, practice size, presence of a female partner, and age and sex structure (table 1). Table 2 compares the morbidity for the Somerset practices with those practices which participated in the fourth national morbidity study.4

Of the 4685 records validated for completeness during April 1994 to March 1995, 4516 (96.4%) of the records were coded. During the same period 1035 (94%) of the 1090 records validated had appropriate episode types and 948 (87%) appropriate Read codes. Of the 142 cases where the Read code was queried, most were not incorrectly coded but were either not coded consistently throughout the course of the illness or not coded in as much detail as was possible from the available information.

The distribution of illness across the disease categories in Somerset was similar to the national distribution (table 2). However, some differences with national data were apparent.

GLUE EAR, ASTHMA, AND HYPERTENSION

Table 3 shows the incidence of glue ear in 0-14 year olds during April 1994 to March 1995 (using all the Read codes which participating general practitioners are currently selecting for glue ear). The incidence of new cases was 17.7 per 1000 0-14 year olds.

Table 4 shows the age and sex breakdown of patients with asthma from Somerset compared with national data. The prevalence of asthma in Somerset is higher than the prevalence nationally for both males and females, with children aged 5-14 years and young adults from 15-24 years having the highest prevalence. Nationally, however, the 0-4 age group has the highest prevalence.4 There was wide variation in prevalence of asthma between participating practices (table 5), practice F having over twice the number of asthmatic patients as practice A after standardising for age and sex.

We found a much higher prevalence of hypertension in Somerset than in the national study (table 4). The variation between practices of the prevalence of hypertension was also large.

Discussion