We used 3 administrative databases to identify anyone in contact with health services in Nova Scotia for psychiatric problems:5
• The Medical Services Insurance database of all fee-for-service claims by NS physicians and psychiatrists, which included patient demographics, date of service, and diagnosis codes from the International Classification of Diseases, Ninth Revision, Clinical Modification
• The Canadian Institute for Heath Information's Discharge Abstract Database, which includes hospital admissions, separation dates, diagnoses and procedures.
• Nova Scotia's Mental Health Outpatient Information System, which records service contacts, demographics and diagnoses in the public sector.
This method is consistent with the definition used by the Public Health Agency of Canada for surveillance of treated psychiatric disorders.7
We will refer to these study subjects in this article as psychiatric patients. The study protocol was approved by the Capital Health Research Ethics Board.
We included all patients whose first psychiatric contact occurred between January 1, 1995 and December 31, 2001 in Nova Scotia. We calculated death rates from the Statistics Canada Vital Statistics Database, using the inception cohort method.4
This reduces survivorship bias by including the entire period at risk. We linked data by using the provincial health card number as a unique identifier. Health card numbers were present in more than 99% of the records, irrespective of the database, and were encrypted to ensure confidentiality.
Administrative data have several advantages over community surveys or data derived from individual clinical settings. They provide accessible longitudinal data for an entire jurisdiction at relatively little cost, and both Health Canada and the Public Health Agency of Canada have used administrative data sets for chronic disease surveillance.7,8
Nevertheless, since these data were designed for billing purposes rather than disease surveillance, studies have measured accuracy over jurisdictions, over time and against other measures.
In reabstraction studies, the Discharge Abstract Database has an accuracy rate of 97%–99% for demographic data, including sex, type of treating physician, dates of hospital admission and discharge, and destination at discharge.9,10
In data from the Mental Health Outpatient Information System, the psychiatric diagnoses recorded showed significant agreement with the relevant items in standardized ratings completed by the treating clinician (Health of the Nation Outcome Scales).11
For instance, patients with schizophrenia and other nonaffective psychoses recorded in the Mental Health Outpatient Information System had significantly higher scores for the Health of the Nation Outcome Scales item measuring delusions and hallucinations (t
test with 410 degrees of freedom [t410
] –10.21), whereas those with depression had the highest ratings for deliberate self-harm (t410
–4.42) and depressive symptoms (t410
–6.32), and patients with anxiety had the highest scores for the Health of the Nation Outcome Scales item where anxiety symptoms are recorded (t410
–3.97; all p
values < 0.001).
Support for the validity of the Medical Services Insurance database comes from agreement across jurisdictions for the prevalance of morbidity with use of physicians' billing data. The prevalence of psychiatric disorders meeting the Public Health Agency of Canada's case definition, about 15%, was similar across Nova Scotia, British Columbia, Alberta and Ontario.7,12
Hospital admissions for these disorders, as captured by the Discharge Abstract Database, accounted for 0.5% of these patients; the remaining 99.5% were determined from billings by provincial physicians. Similarly, the prevalence of diabetes, according to the expanded National Diabetes Surveillance System (a prototype for a comprehensive system of chronic physical disease surveillance, which covers complications such as circulatory disorders and death),8
was constant across Canada at about 5%. The standardized mortality ratios in 1999–2000 for diabetes, from data in the National Diabetes Surveillance System, were 2.10 for Nova Scotian women and 1.70 for NS men, compared with ratios of 2.10 and 1.90, respectively, within Canada's general population.8
Finally, although the administrative databases use ICD-9-CM diagnoses, mental health clinicians make their diagnoses using the Diagnostic and Statistical Manual of Mental Disorders, 4th Edition
Clinicians in publicly funded facilities attend DSM-IV training courses to improve their diagnostic accuracy. All DSM-IV diagnoses have equivalent ICD-9-CM codes.13
The Public Health Agency of Canada's case definition includes ICD-9 diagnoses coded 290 through 319. We also included nonspecific mental disorders outside the formal disorders covered by chapter 5 of ICD-9, such as injury of undetermined intention or psychosocial factors that influence health status, to ensure comparability with previous work4
done in Australia. Because non–chapter 5 diagnoses among patients in primary care would not always be psychiatric, we did not include them.
Where possible, we made use of codes for the ICD-9 diagnosis at hospital discharge. We used a ranking hierarchy of inpatient versus outpatient, and specialist versus primary care, which reflected both increasing percentages of patients with severe mental illness and data reliability. This approach also allowed comparison with the Australian data.4
We grouped disorders into dementia and other organic conditions (290–294), psychoses (schizophrenia or nonaffective psychoses: 295, 297, 299), alcohol or drug disorders (303–305), mood disorders (affective psychoses or depression: 296, 298, 300.4, 311), neuroses (300 except 300.4), personality disorders (301), adjustment reactions (308, 309), and other mental disorders (all remaining chapter 5 and all non–chapter 5 ICD-9 diagnoses of nonspecific mental disorders).
Rates of death were measured for ischemic heart disease (ICD-9 diagnosis codes 410–414), stroke (431–438) and other circulatory diseases (390–459). We calculated age-and-sex–adjusted rates of death by direct standardization, using the average population distributions in Nova Scotia from 1995 through 2001 as the standard weights. Follow-up was judged to begin at the first contact with a clinician and end at death or on December 31, 2001, whichever was earlier. Mortality rate ratios were calculated relative to the rate in the remaining Nova Scotian population.
We calculated standardized rates of first hospital admissions for circulatory disease in a similar way, and used these to calculate admission rate ratios relative to rates in the remaining NS population. We also compared the prevalence of specialized or revascularization procedures among patients receiving psychiatric treatment with those for the remaining general NS population using direct standardization, as previously described. We calculated first-time standardized rates and rate ratios from data from the Discharge Abstract Database for cardiac catheterization, percutaneous transluminal coronary angioplasty (PTCA), coronary artery bypass graft (CABG) or arterial implant, cerebrovascular arteriography, and carotid endarterectomy.
We used proportional hazards or logistic regression, as appropriate, to compare the risk of each outcome (death, first hospital admission or specialized procedure) from time of first contact for psychiatric disorder until the end of follow-up. We included principal psychiatric diagnosis, age (in 10-year increments), sex, socioeconomic status, treatment setting and residence (metropolitan Halifax or elsewhere in the province). We derived income levels using the average household income from the 1996 Census for subjects' postal code at the time of initial contact, and divided them by quartile.
We assessed medical illness leading to hospital admission (as recorded in the Discharge Abstract Database) over the year before study entry by means of the modified Charlson–Deyo index. This contains 19 categories of comorbidity, primarily defined via ICD-9 codes. Each category has an associated weight, based on the adjusted risk of 1-year mortality. The overall score reflects the cumulative increased likelihood of 1-year mortality: the higher the score, the greater the risk.14