Search

Enormous amounts of data are recorded routinely in health care as part of the care process, primarily for managing individual patient care. There are significant opportunities to use these data for other purposes, many of which would contribute to establishing a learning health system. This is particularly true for data recorded in primary care settings, as in many countries, these are the first place patients turn to for most health problems.In this paper, we discuss whether data that are recorded routinely as part of the health care process in primary care are actually fit to use for other purposes such as research and quality of health care indicators, how the original purpose may affect the extent to which the data are fit for another purpose, and the mechanisms behind these effects. In doing so, we want to identify possible sources of bias that are relevant for the use and reuse of these type of data.This paper is based on the authors' experience as users of electronic health records data, as general practitioners, health informatics experts, and health services researchers. It is a product of the discussions they had during the Translational Research and Patient Safety in Europe (TRANSFoRm) project, which was funded by the European Commission and sought to develop, pilot, and evaluate a core information architecture for the learning health system in Europe, based on primary care electronic health records.We first describe the different stages in the processing of electronic health record data, as well as the different purposes for which these data are used. Given the different data processing steps and purposes, we then discuss the possible mechanisms for each individual data processing step that can generate biased outcomes. We identified 13 possible sources of bias. Four of them are related to the organization of a health care system, whereas some are of a more technical nature.There are a substantial number of possible sources of bias; very little is known about the size and direction of their impact. However, anyone that uses or reuses data that were recorded as part of the health care process (such as researchers and clinicians) should be aware of the associated data collection process and environmental influences that can affect the quality of the data. Our stepwise, actor- and purpose-oriented approach may help to identify these possible sources of bias. Unless data quality issues are better understood and unless adequate controls are embedded throughout the data lifecycle, data-driven health care will not live up to its expectations. We need a data quality research agenda to devise the appropriate instruments needed to assess the magnitude of each of the possible sources of bias, and then start measuring their impact. The possible sources of bias described in this paper serve as a starting point for this research agenda.

To determine the appropriate size of risk windows in both exposed and unexposed sub-cohorts.Data was taken from a previous study of upper gastrointestinal haemorrhage and perforation. The length of each prescription for NSAIDs was estimated. The risk was calculated for the duration of a prescription plus increments of -30, -25,..., +115, +120 (i.e. 31 increments). Ten unexposed groups were re-sampled for each increment (stratified for age and sex), using the same lengths of risk window as the exposed group. Mean risks and rate-ratios were calculated (per thousand person-years).The NSAID risk rose from 3.52 at -30 days to a peak of 5.82 at -15 days, and then decreased gradually to 2.83 at +120 days. Unexposed risk was variable for the negative increments, and decreased gradually from 2.16 at +0 days to 1.54 at +120 days. The rate-ratio rose from 1.55 at -30 days to a peak of 2.85 at -5 days, and then decreased to 1.85 at +120 days.Risk windows should be the same as (or slightly less than) the calculated length of a prescription. Lengthy windows should not be used for unexposed comparator groups (the exposed windows may be randomly allocated).

Although widely used, few data are available on the appropriateness of prescribing of acid-suppressing drugs (ASDs), despite guidelines on the investigation and treatment of dyspeptic patients.We created a database of 62 000 endoscopy examinations and record-linked these to a prescribing database. Endoscopic diagnoses were classified into peptic, nonpeptic and others. The H2-antagonists, omeprazole and misoprostol, were studied.35 000 patients had one or more endoscopies during 1978-93; two-thirds were over 45 years of age at first endoscopy. A quarter of all patients who had been endoscoped had consistently normal examinations. Peptic oesophageal pathology was the commonest positive finding. A quarter of those prescribed ASDs between 1989 and 1993 had been endoscoped between 1978 and 1993. In those with a peptic diagnosis prescribed any ASD, the pathologies found were: oesophageal (42.9%), duodenal (36.3%) and gastro-pyloric (21.3%). Patients prescribed omeprazole were more likely to have undergone endoscopy than those prescribed other ASDs, and they were also more likely to have peptic oesophageal pathology. Long-term prescribing (56 days per year) occurred in two-thirds of patients prescribed ASDs and 40% had at least one endoscopy. In those prescribed short-term ASDs, 20% had undergone at least one endoscopy. Peptic and nonpeptic endoscopic pathology was associated with increased ASD prescribing, but a normal endoscopy did not reduce prescribing.ASD prescribing appeared to be mainly symptom-driven. Positive endoscopic findings increased the prescribing of ASDs, but normal findings did not reduce it.

To assess the impact of both general and psychiatric hospitalization on the community prescribing of benzodiazepines, we carried out an observational study using record linkage of prescribing prior to and following hospitalization along with a review of hospital case records at four Tayside General Practices. In a population of 29,672 subjects, 2628 general hospital and 254 psychiatric hospitalizations were studied. The main outcome measure was the change in community benzodiazepine prescribing following hospitalization. We found that admission to a general hospital resulted in 59 of the 2628 subjects (2.2%) commencing and 45 subjects (1.7%) discontinuing benzodiazepines. Admission to a psychiatric hospital resulted in 17 of 254 subjects (6.7%) commencing and 40 (16.7%) discontinuing benzodiazepines. When compared to benzodiazepine prescribing in the study population these effects were trivial. We conclude that hospitalization in both general and psychiatric hospitals had a minor effect on total community prescribing of benzodiazepines. In this study general hospital admission resulted in a small net increase and psychiatric hospitalization a small net decrease in benzodiazepine prescribing.

Few data are available regarding the pattern of prescribing of ulcer-healing drugs (UHDS) in the community, despite their widespread use. The aim of this study was to analyze the pattern of UHD prescribing in Tayside, Scotland (population ≈400 000) between 1989 and 1992.

Scottish Morbidity Record (SMR1) data are coded by trained clerical staff from case records and discharge summaries. They form the basis of many strategic NHS decisions. Their accuracy for upper gastrointestinal (UGI) diagnosis is unknown and the study was undertaken to assess this accuracy in Tayside.Patients who fulfilled the following criteria were identified using a record-linkage pharmacoepidemiological database, and their case records retrieved: over 50 years of age, had encashed at least one prescription for a non-steroidal anti-inflammatory drug at a Tayside pharmacy and who had SMR1 records containing one or more symptom/diagnosis codes between January 1989 and December 1991. Medically qualified staff were trained to examine case records and to code UGI diagnoses. They searched the case records for every UGI SMR1 entry for these patients from 1980-1992 and produced re-coded diagnoses (RCD) for each hospital event (admission and discharge), using all the data available in the case records. They also abstracted data on the clinical presentation, investigations and management of patients. Each event was then examined by a single medically qualified researcher who compared the original SMR1 codes with the RCDs.2,101 patients had a total of 3,764 events in 1989-1991. 317 events were either day case procedures or elective surgery or the case records were not found. They were therefore excluded. Of the remainder, the SMR1 and RCD codes were judged equivalent in 1,608 events (46.6%). However, 1,005 SMR1 events (29.2%) contained a symptom code but no diagnosis code and the remaining 834 (24.2%) were judged suboptimal for other reasons. Of those with a symptom code only, 406 could not be improved upon and were transformed into RCD symptom codes only, 435 were assigned symptom and diagnostic RCDs and 164 were assigned diagnostic RCDs only. In the other 834 events, 279 had one or more diagnoses missing, 425 had one or more diagnoses inaccurate, 23 had both missing and inaccurate diagnoses and 107 were not UGI. Thus 1,433 (41.6%) of UGI SMR1 events could be more accurately coded. Examination of investigation data revealed that coding inaccuracy was not due to diagnostic procedures being carried out after admission.UGI SMR1 data were satisfactory in about half of all events. In about a quarter there were symptom codes but no satisfactory diagnosis codes, whilst in another quarter the data were inaccurate. These findings have implications for health care activities and research that use these data.

Non-steroidal anti-inflammatory drugs (NSAIDs) have been implicated in the aetiology of acute renal failure (ARF), but epidemiological studies examining this association have produced disparate results. We conducted a case-control study using a purpose-built record-linkage database for a population of 420,600 patients, resident in Tayside since May 1990. Patients (n = 207) hospitalized with a diagnostic code for ARF between 1990 and 1992 had their diagnosis validated by a renal physician. Six community controls and two hospital controls, matched for age and sex, were generated for each of these cases. Exposure to dispensed oral NSAIDs, topical NSAIDs and aspirin during the 90 days prior to the index date were investigated (recent exposure), as was exposure at any time since January 1989 (previous exposure). The most significant associations were modelled using conditional logistic regression. When community controls were used, recent exposure to NSAIDs and previous exposure to aspirin were independently associated with hospitalization for ARF, with adjusted odds ratios of 2.20 (1.49-3.25) and 2.19 (1.46-3.30), respectively. Only recent exposure to oral NSAIDs was associated when hospital controls were used: 1.84 (1.14-2.93). No significant interactions were present with previous chronic renal failure, other possible causes of ARF or whether the diagnosis was primary or secondary. There is an approximate doubling of the risk of hospitalization for ARF with use of oral NSAIDs.

This was an observational study of efficacy and resource utilisation over three years in a cohort of 917 patients who received at least one prescription for an oral antibiotic between January and March 1989 in a large rural primary care health centre in Tayside, Scotland. Three thousand, six hundred and sixty three prescriptions were issued (2286 to females) for; 1502 upper respiratory tract infections, 419 lower respiratory tract infections, 441 urinary tract infections, 177 skin and soft tissue infections, 97 gynaecological infections, 103 cases of acne and 71 other infections. Excluding acne, 14% of infections required more than one antibiotic to achieve a successful outcome. The highest success rates for antibiotics were amoxycillin (92%) and penicillin V (92%) in upper respiratory tract infections, erythromycin (90%) in lower respiratory tract infections and co-trimoxazole (83%) in urinary tract infections. The most cost-effective antibiotics were penicillin V for upper respiratory tract infection, erythromycin for lower respiratory tract infection and co-trimoxazole for urinary tract infection. Varying the value placed on general practitioners' time did not change the rank order of antibiotic cost-effectiveness. The efficacy and cost-effectiveness of the treatment of acute infections in primary care varies considerably with the antibiotic used as first choice. More expensive antibiotics might be justified in cost-effectiveness terms if they had high cure rates in clinical practice.

It is generally accepted that antibiotic use can result in vaginal fungal overgrowth, although evidence estimating the extent to which this causes symptomatic vaginitis is scant. In a study using the prescription of vaginal antifungal preparations as a surrogate measure of vaginal candidiasis, a cohort of women taking antibiotics had a higher incidence of vaginal candidiasis after antibiotic exposure than beforehand (relative risk 2.3; 95% confidence interval 1.9-3.0); this risk was highest in those aged 36-40 years (RR 6.0, 95% CI 2.9-12.5). The attributable risk was highest among those who were taking cephalosporins (AR 12.8%, 95% CI 9.1-16.5). In a case-control study, comparing previous antibiotic exposure among women using vaginal antifungal agents and matched controls, antibiotic exposure was higher among those using vaginal antifungal agents during the previous 28 days, with an odds ratio of 5.5 (95% CI 3.8-7.9).