Abstract
Record-linkage is the linkage of patient-specific information that is stored separately. Recent advances in computerization have meant that record-linkage techniques in medical research are increasingly being used and refined. In particular, they have made a significant contribution to pharmacovigilance, which involves linking drug exposure to outcomes data. In this article, the contribution of record-linkage in Scotland to medical research is described. The two organizations that utilize record-linkage techniques are the Medicines Monitoring Unit (MEMO) of the University of Dundee and the Information and Statistics Division (ISD) of the NHS in Scotland. Pharmacovigilance is MEMO’s main concern (using data from the Tayside region of Scotland), while ISD link health care datasets for Scotland for general health care research. The experience of the two groups is now being combined to carry out drug safety studies in the entire population of Scotland.
Keywords: MEMO, pharmacovigilance, record-linkage
Introduction
The term record-linkage was coined by Dunn in 1946 [1] who used it to designate the linking of various records of a person’s life, with this analogy; ‘Each person in the world creates a Book of Life. This book starts with birth and ends with death. Its pages are made of records of the principal events in life. Record linkage is the name given to the process of assembling the pages of this book into a volume’. In other words, record-linkage is the systematic bringing together of the records of individuals in a large population. This is particularly important for medical records, as medical data arising from different health care contacts are usually stored separately. By linking all the medical information about individuals in a population, even where this information originates in different data sources, all medical events can be considered together. Record-linkage has been facilitated in recent years by advances in computerization.
There are two methods by which the linkage of records, belonging to the same individual but stored separately, can be achieved. Deterministic linkage refers to linkage that is based on the use of defined identifiers for each individual. These are usually assigned centrally and used in any records that are kept for that individual. Alternatively, probabilistic linkage is based on combinations of non-unique characteristics of each individual, such as name, date of birth or gender. These standard identifying items are prone to discrepancies due to error or changes in status. Therefore patterns of agreement and disagreement between identifying items are translated into quantitative scores, which are then used to predict whether the two records should be linked.
Currently, there are two major record-linkage initiatives in Scotland. The Medicines Monitoring Unit (MEMO) was set up at the University of Dundee to carry out post marketing drug safety studies and therefore has a direct role in pharmacovigilance. The Information and Statistics Division (ISD) of the NHS Common Services Agency in Edinburgh also works on record-linkage, although so far their main contribution to pharmacovigilance has been collaboratively with MEMO. The work of both groups is described in this article.
The Medicines Monitoring Unit (MEMO)
Record-linkage at MEMO
MEMO is a university-based research organization that record-links data for the population of the Tayside region of Scotland. This is enabled by the widespread use of the Community Health Index Number (CHNo) in Tayside, the potential benefits of which were recognized in the 1980s [2]. The CHNo is a 10-digit number allocated to all patients when they register with a general practitioner (GP) in Scotland. The number is unique and specific to each patient, with the first six digits representing date of birth, then a three digit serial number, the last digit of which indicates sex, and a finally a check sum digit to ensure the validity of the CHNo. If a patient is assigned more than one CHNo in error, these can then be linked together. Although all patients registered with a GP in Scotland officially have a CHNo, Tayside Health Board is the only one that uses the CHNo routinely as a patient identifier. The Community Health Index is a computerised list of the CHNos of all GP-registered patients, containing names and addresses and up-to-date demographic details. Tayside has a population of ≈400 000 people (the estimated mid-year resident population was 395 600 in 1995 [3]). The demographic breakdown is broadly similar to that of the rest of Scotland, as is the general health status of the population.
Record-linkage in MEMO has both a deterministic and probabilistic element; deterministic because the Community Health Index is validated and maintained centrally by Tayside Health Board, and MEMO makes use of many datasets that are already indexed by the CHNo. However, to exploit databases that are patient-specific, but do not use the CHNo as a patient-identifier, it is possible to match patient records to the correct CHNo, by comparing non-unique characteristics of the patient to those held with the CHNo in the Community Health Index. This is usually carried out within MEMO.
Pharmacovigilance
MEMO’s role in pharmacovigilance is the postmarketing surveillance (PMS) of drugs, (referring to any nonexperimental method used for assessing side-effects of drugs, whether detrimental or beneficial, after marketing). The use of record-linkage databases, linking computerised drug exposure data to medical outcomes in large populations, has made a significant contribution to this field in recent years, particularly in North America [4], and it was for this purpose that the record-linkage database was originally constructed at MEMO [5, 6].
The pivotal datasets are an ‘exposure’ database of dispensed prescriptions, and an ‘outcome’ database of hospital admissions. The exposure database is compiled from copies of all prescriptions that have been dispensed in community pharmacies in Tayside, and includes all drug and patient details recorded on the form. The CHNo is allocated to the records routinely by MEMO. Constructing the database is an ongoing concern, but it currently holds records of 15 million prescription items dispensed since 1989 (all drugs dispensed since January 1993 and selected drugs prior to this). Tayside Scottish Morbidity Record 1 (SMR1) is the outcome dataset of hospital admissions, supplied to MEMO every year by ISD. For every consultant episode of care, there are demographic data, such as age, sex and postcode of the patient, and administrative data, such as date of admission, date of discharge and length of stay. The clinical information is in the form of up to six ICD9 diagnosis codes, from the International Classification of Diseases, ninth revision [7], and up to four OPCS4 operation or procedure codes, from the fourth revision of the Office of Population Censuses and Surveys [8]. The Tayside SMR1 database also routinely contains the CHNo for every patient.
Other ISD datasets, for example SMR2 (maternity admissions), SMR4 (mental health admissions) and SMR11 (neonatal admissions), are also used in MEMO. Indeed, any health care dataset that is indexed by the CHNo can be linked to MEMO’s exposure database, as can any dataset that contains some patient-specific demographic detail, such as name, date of birth and postcode, from which the CHNo can be identified. For example, SMR6 (the Cancer Registration database) and results from biochemical tests carried out in Tayside hospitals have been linked in this way. Some outcome databases have also been constructed by MEMO from scratch, notably databases of 82 000 endoscopy and colonoscopy procedures, and victims of 19 000 road traffic accidents in Tayside.
In the last few years, many studies have linked outcome databases to the dispensed prescribing database in MEMO to quantify the risks associated with prescribed drugs. Thus far, MEMO’s role has mainly been in hypothesis-testing, that is investigating drug hazards flagged up by other methods of PMS, such as spontaneous reporting. For drugs that have been on the market for a long time, MEMO is unlikely to have a prominent hypothesis-generating role (that is, identifying previously unsuspected risks), but could certainly monitor the safety of newly marketed drugs.
The risks of nonsteroidal anti-inflammatory drugs (NSAIDs) have been investigated quite extensively in MEMO. That there is a risk of upper gastrointestinal haemorrhage and perforation associated with NSAID use is well known [9], but MEMO has described the risk profiles in more detail, suggesting that the risks are constant with continuous exposure [10] (in contrast to previously published findings [11]), and that they are confined to patients without a previous history of upper gastrointestinal events [12]. MEMO also showed that the risk is with oral NSAIDs, not topical NSAIDs [13].
In addition MEMO has provided evidence for other suspected risks associated with oral NSAID use, notably acute renal failure [14] and colitis [15]. However, pharmacovigilance is also concerned with refuting false safety scares so that effective drugs are not removed from the market unnecessarily. For example, a case-control study showed that there was no association between NSAIDs and acute appendicitis [16], and discussed reasons why an earlier study had found a six-fold risk [17].
An early study in MEMO investigated the risks of vaginal candidiasis associated with antibiotic use using a plotting technique [18], and this work is now being extended to look at categories of antibiotics separately. Using cost data that is incorporated in the prescribing database, MEMO is also able to carry out pharmacoeconomic analyses, for example with antibiotics [19]. Other pharmacovigilance issues that MEMO is now exploring include the risks of road traffic accidents associated with prescribed drug use, and the safety of nabumetone and anorectic drugs.
Audit of prescribing
An important dimension of pharmacovigilance in MEMO is the audit of prescribing in the population, although GP-specific data are analysed anonymously and individual GPs are never identified. For example, one study identified rare instances of potentially hazardous coprescribing of β-adrenoceptor antagonists and β-adrenoceptor agonists to patients in Tayside likely to have asthma or chronic obstructive airways disease [20]. Prior to publication, the results of the study were circulated to all GPs in Tayside, and those that requested information on their patients were supplied with the relevant study data. This is a good example of how MEMO can work with GPs to promote safer prescribing practices. A similar approach was taken in a study that investigated antidepressant prescribing in the population, showing that these drugs were often prescribed at ineffective doses for insufficient durations [21]. There is now a need to repeat the studies to determine whether publication of the results has changed GP prescribing behaviour, thereby completing the audit cycle. The processing of prescribing data according to the demographic characteristics of prescribing GPs has also yielded some useful insights into the characteristics of ‘good’ prescribers. For example, differences in the prescribing of antibiotics and psychotropic medication have been seen between GP registrar training and nontraining practices [22]. Indeed, a potentially interesting avenue of research might be the evaluation of educational interventions in GP practice using the dispensed prescribing database.
Prescribing might also vary by patient factors, independent of need or disease severity, a classic example being variation in the use of hormone replacement therapy by social class [23]. A related issue is patient compliance to medication, as drug therapy will be ineffective if it is not taken properly. By assessing how medication is collected by patients, in terms of numbers of prescriptions dispensed and intervals between them, and linking this to outcome datasets, it is possible to assess the effects of primary non-compliance. This is underway for asthma and heart failure, and has been completed in diabetes, showing that adolescents in Tayside who have ‘brittle’ diabetes are often non-compliant with insulin [24].
Disease registers and the DARTS project
An interesting application of record-linkage is the compilation of patient-specific disease registers using information from different sources. In Tayside, a disease register of all patients (treated or nontreated) with type 1 and type 2 diabetes has been constructed. This is known as the DARTS initiative (Diabetes Audit and Research in Tayside Scotland), a collaboration between MEMO, all GP practices in Tayside and the Diabetes Units in three Health Care Trusts [25]. A disease register uses all relevant information on patients in a defined population to identify those with a given disease in order to provide reliable estimates of the geographical incidence of disease, to facilitate observational studies on its aetiology and natural history, and to calculate indirect costs to individuals and society.
The register is compiled by record-linking data from eight independent data sources [25]; Scottish Morbidity Record 1 (SMR1), four diabetes clinics in three Health Care trusts, and a mobile eye van that has been operating in Tayside since 1990 performing community retinopathy screening, all of which use the CHNo routinely as a patient identifier. Data from MEMO’s dispensed prescribing database (which includes glucose monitoring equipment) and the results of biochemistry tests in hospitals are also used. Because the CHNo is assigned by MEMO to these records on the basis of patient characteristics (names, date of birth, GP), DARTS has a probabilistic linkage component.
The ascertainment of diabetes by DARTS has been validated [25]. Internal validation of the computerised data is essential, and incorporated with this work is ongoing assessment of the utility and accuracy, in terms of sensitivity and specificity, of the separate data sources for identifying diabetic patients. The other dimension of validation is the checking of the primary medical records (GP records) of patients identified by DARTS as diabetic. This work suggests that record-linkage is a more sensitive method for identifying diabetic patients than the use of GP registers, with sensitivity values of 96% and 91%, respectively [25].
DARTS is to be used to investigate the epidemiology of diabetes (with some research already completed [24, 26, 27]), and for health care research, audit and patient management. Similar methods are being explored for cardiovascular disease, epilepsy, schizophrenia and depression. While they could also be adopted by other regions in Scotland, the well-laid foundations for record-linkage that already existed in Tayside, and extensive experience with these systems, proved to be essential to the DARTS initiative.
The Information and Statistics Division
The Information and Statistics Division (ISD) is the central hub of data collection on health care for the population of Scotland. Here data are collected, stored and analysed in order to develop and to evaluate health care and health policy, to monitor health outcomes and the health status of the population, and to allocate resources most effectively. For example, Scottish Morbidity Records (SMR) are sets of patient-specific data on health care activity in Scotland, and include outpatient attendances (SMR0), general inpatient and day cases (SMR1), maternity admissions (SMR2), mental health admissions (SMR4), neonatal admissions (SMR11) and geriatric admissions (SMR50). Other SMR record types include waiting lists (SMR3), cancer registration (SMR6), community dental services (SMR13), cardiac surgery (SMR20), drug misuse (SMR22/23) and accident and emergency waiting times (SMR30C). In addition, ISD holds data on population and vital statistics (including birth records, death records with cause of death from the Registrar General, abortions, stillbirths and neonatal deaths), family planning and genito-urinary medicine services (including HIV), immunization records, notifications of infectious diseases and screening services. Reference to any ISD publication would provide a fuller listing of the datasets available [28].
Record-linkage at ISD
ISD uses probability matching to link selected datasets. The core items of information used are surname, initial, and year, month and day of birth. To establish whether pairs of records should be matched, or linked together, a computer algorithm calculates a score that is proportional to the likelihood that they belong to the same individual. The overall score is derived from a comparison of each item of identifying information, and has an estimated accuracy of 98–99%, although this often depends on the number of patient records being linked into a patient dataset. The probability that a certain record does not belong increases with the size of the patient dataset, and it is therefore important to have some inbuilt error checking. The ISD linkage methods have been described in detail [29], and more recently ‘one-pass linkage’ techniques have been developed [30]. This increases the speed and efficiency by which small datasets can be linked to larger ones, by storing the small dataset in memory and affording direct access to these records.
With a population of 5.5 million in Scotland (the estimated mid-year resident population was 5 136 600 in 1995 [3]), record-linkage of selected datasets is both manageable and informative. For example, SMR1, cancer registration, and death records of 3.5 million patients, representing 11 million hospital contacts, have been linked for the period 1981–95. SMR4 (mental health admissions) and death records have been linked from 1970 onwards. Maternity, neonatal, stillbirth, infant death and birth records have been linked since 1980. Until now, the record-linked databases have mainly been used for evaluation of health care activity [31] and outcomes research [32]. However, research possibilities in pharmacovigilance are now being explored in a collaborative venture with MEMO.
Pharmacovigilance in Scotland: collaboration between MEMO and ISD
The most exciting record-linkage initiative in Scotland is probably the piloting of drug safety studies in the entire population of Scotland [33]. While MEMO’s record-linkage database has made a significant contribution to the evaluation of drug safety, the underlying population is too small for accurate quantification of the risks of adverse drug reactions that occur at low frequency in the population (either because the drug is rarely prescribed or because the reaction itself is rare). This collaborative project therefore draws upon MEMO’s expertise in pharmacovigilance and ISD’s experience in probabilistic matching, and the Pharmacy Practice Division provide the raw exposure data (facsimiles of prescriptions). Three test drugs have been chosen, azapropazone, dornase alpha and losartan. MEMO has obtained facsimiles of prescriptions dispensed in Scotland between April 1994 and August 1995 for these drugs, and allocated CHNos to them using Scottish-wide CHNo databases. On the basis of up-to-date demographic details provided by the CHNo, ISD has used a probability match on sex, full name, date of birth and current and previous postcodes to identify any SMR1 hospital admission data from 1981 to June 1995 for the 14 536 patients who received these drugs. There were 43 457 hospital episodes for 10 248 of them. Preliminary analyses show that the rate of hospital admission for upper gastrointestinal haemorrhage and perforation was 11.3 events per 1000 patient years for patients exposed to azapropazone, compared with 2.0 per 1000 patient years for unexposed patients, giving an incidence rate ratio of 5.7 (95% confidence intervals 3.5–9.3). That these results fit in with what is already known about the risk profile of azapropazone is encouraging. Analyses for the other drugs are underway. Now that this pilot study has demonstrated that patient-specific exposure datasets can be compiled for the Scottish population, the range and quality of the outcome data to which they can be linked by ISD would make drug safety studies using these resources unrivalled in detail and scope. These possibilities are still being explored.
Confidentiality and ethics in record-linkage
The types of record-linkage projects described here involve highly confidential medical data, and organizations that carry out record-linkage usually adhere to strict conditions. For example, MEMO has an agreement with the Local Medical Committee of the British Medical Association never to divulge person-specific or GP-specific data, unless it is to a doctor requesting information on one of his or her own patients. All staff in MEMO sign confidentiality agreements and all databases are registered for research purposes with the Data Protection Officer. All studies use data that are anonymised by the CHNo.
With regard to independent ethical approval for drug safety studies, guidelines issued by the Scottish Office indicated that post marketing surveillance studies were exempt from this requirement [34]. Research involving access to medical records without direct patient involvement is also exempt provided that confidentiality and anonymity of patients is assured, and a senior professional person who can be disciplined by a professional body is accountable. The rationale for this is set out in a recent report into ethical issues from a Working Group of the Royal College of Physicians of London [35]. Although it is clear that epidemiologists take issues of confidentiality very seriously [36], recently there have been some worrying developments. For example, suggested EU guidelines on restricting the use of data for purposes for which they were originally gathered, and obtaining informed consent from each person whose data are used, will make epidemiological studies, and in particular record-linkage studies, impossible [37]. A proposed solution is the encoding of data by third parties, and making the anonymised data available to researchers [38]. However, this may not be sufficient to protect drug safety research.
With the potential of record-linkage now going beyond pharmacovigilance, and facilitating outcomes research, general epidemiological studies and even economic evaluation, it is essential that a workable solution is reached. The contribution that record-linkage can make to medical research is invaluable, and it would be disappointing if this were obstructed before the value of record-linkage is fully appreciated.
References
- 1.Dunn H. Record linkage. JAMA. 1946;36:1412–1416. [PubMed] [Google Scholar]
- 2.Crombie IK, Brown SV, Hamley JG. Postmarketing drug surveillance by record linkage in Tayside. J Epidemiol Community Health. 1984;38:226–231. doi: 10.1136/jech.38.3.226. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Information & Statistics Division. Scottish Health Statistics 1996. Edinburgh: ISD Scotland Publications; 1996. [Google Scholar]
- 4.Strom BL, Carson JL. Use of automated databases for pharmacoepidemiology research. Epidemiol Rev. 1990;12:87–107. doi: 10.1093/oxfordjournals.epirev.a036064. [DOI] [PubMed] [Google Scholar]
- 5.MacDonald TM, McDevitt DG. The Tayside Medicines Monitoring Unit (MEMO) In: Strom BL, editor. Pharmacoepidemiology. 2. Chichester: John Wiley & Sons Ltd; 1994. pp. 245–255. [Google Scholar]
- 6.Evans JMM, McDevitt DG, MacDonald TM. The Tayside Medicines Monitoring Unit (MEMO): A record-linkage system for pharmacovigilance. Pharmaceut Med. 1995;9:177–184. [Google Scholar]
- 7.International Classification of Diseases. Manual of the International Statistical Classification of Diseases, Injuries and Causes of Death. Vol. 1. Geneva: World Health Organisation 1977; 1977. Ninth revision. [Google Scholar]
- 8.Office of Population Censuses and Surveys. Tabular list of the classification of surgical operations and procedures, fourth revision. HMSO; 1990. [Google Scholar]
- 9.Hawkey CJ. Non-steroidal anti-inflammatory drugs and peptic ulcers. Facts and figures multiply, but do they add up? Br Med J. 1990;300:278–284. doi: 10.1136/bmj.300.6720.278. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.MacDonald TM, Morant SV, Robinson GC, et al. The upper gastrointestinal toxicity of non-steroidal anti- inflammatory drugs is constant with continued exposure. Br Med J. 1997;315:1333–1337. doi: 10.1136/bmj.315.7119.1333. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Carson JL, Strom BL, Soper KA, West SL, Morse ML. The association of nonsteroidal anti-inflammatory drugs with upper gastrointestinal tract bleeding. Arch Intern Med. 1987;147:85–88. [PubMed] [Google Scholar]
- 12.McMahon AD, Evans JMM, White G, et al. A cohort study (with resampled comparator groups) to measure the association between new NSAID prescribing and upper gastrointestinal haemorrhage and perforation. J Clin Epidemiol. 1997;50:351–356. doi: 10.1016/s0895-4356(96)00361-7. [DOI] [PubMed] [Google Scholar]
- 13.Evans JMM, McMahon AD, McGilchrist MM, et al. Topical non-steroidal anti-inflammatory drugs and admission to hospital for upper gastrointestinal bleeding and perforation: a record-linkage case- control study. Br Med J. 1995;311:22–26. doi: 10.1136/bmj.311.6996.22. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Evans JMM, McGregor E, McMahon AD, et al. Non-steroidal anti-inflammatory drugs and hospitalization for acute renal failure. Q J Med. 1995;88:551–557. [PubMed] [Google Scholar]
- 15.Evans JMM, McMahon AD, Murray FE, McDevitt DG, MacDonald TM. Non-steroidal anti- inflammatory drugs are associated with emergency hospitalisation for colitis due to inflammatory bowel disease. Gut. 1997;40:619–622. doi: 10.1136/gut.40.5.619. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Evans JMM, MacGregor A, Murray FE, Vaidya K, Morris AD, MacDonald TM. No association between non-steroidal anti-inflammatory drugs and acute appendicitis in a case-control study. Br J Surg. 1997;84:372–374. [PubMed] [Google Scholar]
- 17.Campbell KL, De Beaux AC. Non-steroidal anti-inflammatory drugs and appendicitis in patients aged over 50 years. Br J Surg. 1992;79:967–968. doi: 10.1002/bjs.1800790938. [DOI] [PubMed] [Google Scholar]
- 18.MacDonald TM, Beardon PHG, McGilchrist MM, Duncan ID, McKendrick AD, McDevitt DG. The risks of symptomatic vaginal candidiasis after oral antibiotic therapy. Q J Med. 1993;86:419–424. [PubMed] [Google Scholar]
- 19.MacDonald TM, Collins D, McGilchrist MM, et al. The utilisation and economic evaluation of antibiotics prescribed in primary care. J Antimicrob Chemother. 1995;35:191–204. doi: 10.1093/jac/35.1.191. [DOI] [PubMed] [Google Scholar]
- 20.Hayes JL, Evans JMM, Lipworth BP, MacDonald TM. Potentially hazardous co- prescribing of β-adrenoceptor antagonists and agonists in the community. Br J Gen Pract. 1996;46:423–425. [PMC free article] [PubMed] [Google Scholar]
- 21.MacDonald TM, McMahon AD, Reid IC, Fenton GW, McDevitt DG. Anti-depressant drug use in primary care: A cause for concern. Br Med J. 1996;313:860–861. doi: 10.1136/bmj.313.7061.860. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Steinke DT, Bain DJG, Davey PG, MacDonald TM. Practice factors that influence prescribing of antibiotic and psychotropic medication in general practice in Tayside. Br J Gen Pract. (Submitted for publication)
- 23.Evans JMM, Orr C, Duncan ID, MacDonald TM. Prescribing of hormone replacement therapy in the community: Could this be improved? Pharmacoepidemiol Drug Safety. 1997;6:81. [Google Scholar]
- 24.Morris AD, Boyle DIR, McMahon AD, Greene SA, MacDonald TM, Newton RW. Adherence to insulin treatment, glycaemic control and ketoacidosis in insulin dependent diabetes mellitus. Lancet. 1997;350:1505–1510. doi: 10.1016/s0140-6736(97)06234-x. [DOI] [PubMed] [Google Scholar]
- 25.Morris AD, Boyle DIR, MacAlpine R, et al. The diabetes audit and research in Tayside Scotland (DARTS) study: electronic record-linkage to create a diabetes register. Br Med J. 1997;315:524–528. doi: 10.1136/bmj.315.7107.524. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Morris AD, Boyle DIR, McMahon AD, et al. ACE inhibitor use is associated with hospitalization for severe hypoglycemia in patients with diabetes. Diabetes Care. 1997;20:1363–1367. doi: 10.2337/diacare.20.9.1363. [DOI] [PubMed] [Google Scholar]
- 27.Morris AD, Foster G, Boyle DIR, et al. Diabetes associated lower extremity amputation in the community: are the St Vincent Declaration targets achievable? Diabetic Med. 1996;13:23. [Google Scholar]
- 28.Information and Statistics Division. ISD Scotland Guide: an A–Z of the work of the Information & Statistics Division. Edinburgh: Information & Statistics Division; 1996. [Google Scholar]
- 29.Kendrick S, Clarke J. The Scottish record-linkage system. Health Bull (Edinb) 1993;51:72–79. [PubMed] [Google Scholar]
- 30.Kendrick S, McIlroy R. One-pass linkage—rapid creation of patient-based data. Healthcare computing. 1996:589–598. [Google Scholar]
- 31.Kendrick S. The pattern of increase in emergency hospital admissions in Scotland. Health Bull (Edinb) 1996;54:169–183. [PubMed] [Google Scholar]
- 32.Clinical Outcomes Working Group. Clinical Outcome Indicators. Edinburgh: Information & Statistics Division; 1996. p. 1996. [Google Scholar]
- 33.McGilchrist MM, McMahon AD, MacDonald TM. A study to assess record-linkage in the Scottish population. Pharmacoepidemiol Drug Safety. 1997;6:164. [Google Scholar]
- 34.The Scottish Office, Home and Health Department. Local Research Ethics Committees. HMSO; [Google Scholar]
- 35.Diamond A, Hall S, Jay M, Laurence D, et al. Independent ethical review of studies involving personal medical records. Report of a working group to the Royal College of Physicians Committee on Ethical Issues in Medicine. J R Coll Phys Lond. 1994;28:439–442. [PMC free article] [PubMed] [Google Scholar]
- 36.Last J. Workshop: Epidemiology and ethics. Lancet. 1990;336:497. [Google Scholar]
- 37.Editorial The ethics of learning from patients. Lancet. 1994;344:71–72. [PubMed] [Google Scholar]
- 38.Horner JS. Computers can be compatible with confidentiality. J Roy Coll Phys Lond. 1997;31:310–312. [PMC free article] [PubMed] [Google Scholar]