The association of objectively ascertained sibling fracture history with major osteoporotic fractures: a population-based cohort study.

We investigated the association of objectively ascertained sibling fracture history with major osteoporotic fracture (hip, forearm, humerus, or clinical spine) risk in a population-based cohort using administrative databases. Sibling fracture history is associated with increased major osteoporotic fracture risk, which has implications for fracture risk prediction. INTRODUCTION: We aimed to determine whether objectively ascertained sibling fracture history is associated with major osteoporotic fracture (MOF; hip, forearm, humerus, or clinical spine) risk. METHODS: This retrospective cohort study used administrative databases from the province of Manitoba, Canada, which has a universal healthcare system. The cohort included men and women 40+ years between 1997 and 2015 with linkage to at least one sibling. The exposure was sibling MOF diagnosis occurring after age 40 years and prior to the outcome. The outcome was incident MOF identified in hospital and physician records using established case definitions. A multivariable Cox proportional hazards regression model was used to estimate the risk of MOF after adjustment for known fracture risk factors. RESULTS: The cohort included 217,527 individuals; 91.9% were linked to full siblings (siblings having the same father and mother) and 49.0% were females. By the end of the study period, 6255 (2.9%) of the siblings had a MOF. During a median follow-up of 11 years (IQR 5-15), 5235 (2.4%) incident MOF were identified in the study cohort, including 234 hip fractures. Sibling MOF history was associated with an increased risk of MOF (hazard ratio [HR] 1.67, 95% confidence interval [CI] 1.44-1.92). The risk was elevated in both men (HR 1.57, 95% CI 1.24-1.98) and women (HR 1.74, 95% CI 1.45-2.08). The highest risk was associated with a sibling diagnosis of forearm fracture (HR 1.81, 95% CI 1.53-2.15). CONCLUSION: Sibling fracture history is associated with increased MOF risk and should be considered as a candidate risk factor for improving fracture risk prediction.

Effect of Study Duration and Outcome Measurement Frequency on Estimates of Change for Longitudinal Cohort Studies in Routinely-Collected Administrative Data.

INTRODUCTION: When designing longitudinal cohort studies, investigators must make decisions about study duration (i.e. length of follow-up) and frequency of outcome measurement. This research explores these design decisions for longitudinal cohort studies constructed using routinely-collected administrative data. OBJECTIVES: To illustrate the effects of varying study duration and frequency of outcome measurement in longitudinal cohort studies conducted using routinely-collected administrative data using a numeric example. METHODS: Linked administrative data from Manitoba, Canada were used. The cohort included mothers who experienced the death of an infant between April 1, 1999 and March 31, 2012 and a matched (three:one) group of mothers who did not experience an infant death. A generalized linear model was used to test for differences between groups in the non-linear (i.e. quadratic) and linear trend over time for the number of healthcare contacts. Holding sample size constant, models were fit to the data for various combinations of study duration and measurement frequency. Regression coefficient estimates and their standard errors were compared. RESULTS: A total of 2576 mothers were included; 644 experienced an infant death and 1932 were matches. Thirteen combinations of measurement frequency (one, two, three, four periods/year) and study duration (one, two, three, four years) were investigated. As frequency increased from one to four periods/year, the standard errors of the regression coefficients for the group difference in the non-linear trend (i.e. group-time-time interaction) decreased up to 98.9%. As duration increased from one to fours years, the standard errors decreased up to 96.9%. As frequency and duration increased, the estimated regression coefficients trended toward zero. Similar results were observed for the linear trend model. CONCLUSION: Longitudinal cohort studies based on administrative data offer flexibility in time-related design elements, but present potential challenges. Recommendations about how to select and report design decisions in studies should be included in reporting guidelines.

The Concept Dictionary and Glossary at MCHP: Tools and Techniques to Support a Population Research Data Repository.

The Manitoba Centre for Health Policy's Concept Dictionary and Glossary, and the Data Repository they document, broaden the analytic possibilities associated with administrative data. The aim of the Repository is to describe and explain patterns of health care and illness, while the Concept Dictionary and Glossary create consistency in documenting research methodologies. The Concept Dictionary alone contains detailed operational definitions and programming code for measures used in MCHP research that are reusable in future projects. Making these tools available on the internet allows reaching a heterogeneous audience of academic and government health service partners, epidemiologists, planners, programmers, clinicians, and students extending around the globe. They aid in the retention of corporate knowledge, facilitate researcher/analyst communication, and enhance the Centre's knowledge translation activities. Such documentation has saved countless hours for programmers, analysts and researchers who frequently need to tread paths previously taken by others.

A research registry: uses, development, and accuracy.

Many studies of population health, clinical epidemiology, and health services can be supported by a population-based research registry. Such a registry accurately defines the health insurance status for each individual over many years, magnifying the effectiveness of a cross-sectional registry (typically relevant for only a short duration) used in the administration of a health insurance plan. A research registry can distinguish between "well" individuals (no contact with the health care system), loss to follow-up (ineligibility associated with leaving the insurance plan), loss of continuity (two or more unlinked registrations over time for the same person), and mortality. The Manitoba research registry was developed to facilitate longitudinal studies; working within strict confidentiality controls, identifiers for each individual known to Manitoba Health since 1970 can be retrieved and a single unique identifier assigned. Careful reporting of changes in family registration numbers has enabled tracing area of residence, marital status, and family characteristics; results are equivalent to a daily census of the province. This article provides details on source materials, design, and quality of the registry, highlighting its value both for the development of integrated population health information systems and for research in general.

Risk adjustment in claims-based research: the search for efficient approaches.

Claims-based indices of comorbidity and severity, as well as other measures derived from routinely collected administrative data, are developed and tested. The extent to which risk adjustments using claims can be improved by adding information from one well-known measure based on chart review and patient examination (the American Society of Anesthesiologists' (ASA) Physical Status score) is also examined. Readmissions and mortality after three common surgical procedures are the outcomes studied using multiple logistic regression. Claims-based measures of comorbidity, derived both from hospital discharge abstracts at the time of surgery and from hospitalizations in the 6 months before surgery, provided reasonably good predictions of postsurgical readmissions and mortality. In the most complete logistic regression models, the Somers' Dyx measure of fit (a rank correlation coefficient) ranged from 0.23 to 0.38 for readmissions and from 0.46 to 0.72 for mortality. In 5 out of 6 cases, these predictions were not improved by including the prospectively-collected ASA Physical Status score. Such difficulties in improving risk adjustment by more intensive data collection are discussed in terms of their research implications.

Comparing clinical information with claims data: some similarities and differences.

How well can hospital discharge abstracts be used to estimate patient health status? This paper compares information on comorbidity obtained from hospital discharge abstracts for patients undergoing prostatectomy or cholecystectomy at a Winnipeg teaching hospital with clinical data on preoperative medical conditions prospectively collected during an Anesthesia Follow-up study. The diagnostic information on cardiovascular disease, respiratory disease, and metabolic disorders showed considerable agreement, ranging from 65 to over 90% correspondence across the two data sets. Certain conditions noted by the anesthesiologist were often absent from the claims data; cardiovascular disease was recorded in the clinical data but absent from the claims for 31% of prostatectomy and 17% of cholecystectomy cases. Such patients were less likely to have been assigned a high score on the ASA Physical Status measure or to have high-risk diagnoses on the hospital file. Similar findings resulted from comparing the two sources in their ability to predict such adverse outcomes as mortality and readmission to hospital: the anesthesia file generally included less serious comorbidity.

Comparison of survey and physician claims data for detecting hypertension.

Using linked data from the Manitoba (Canada) Heart Health Survey (MHHS) and physician service claims files we assessed the degree to which self-reported hypertension and clinically measured hypetension agreed with physician claims hypertension, and examined the likely sources of disagreement. The overall agreement between survey and claims data for hypertension detection was moderate to high: 82% (kappa = 0.56) for self-reported and physician claims hypertension, and 85% (kappa = 0.60) for clinically measured and physician claims hypertension. In the comparison between self-report and physician claims, those who were classified as obese, diabetic, or a homemaker were significantly more likely to have a hypertension measure not confirmed by the other. Disagreement between clinically measured and physician claims was also more common among the obese and homemakers, as well as those on medication for heart diseases, elevated cholesterol levels (LDL), and 35 years of age and older. The high overall level of agreement among these three measures suggest that each may be used with confidence as an indication of hypertension; however, the agreement appears lower among individuals presenting a more complicated clinical profile.

A comparison of administrative versus clinical data: coronary artery bypass surgery as an example. Ischemic Heart Disease Patient Outcomes Research Team.

Health services researchers rely heavily on administrative data bases, but incomplete or incorrect coding may bias risk models based on administrative data. The best method for validating administrative data is to collect detailed information about the same cases from independent sources, but this approach may be too costly or technically difficult. We used data on coronary artery bypass surgery from four sites (Duke University; Minneapolis--St Paul; California; and Manitoba) to demonstrate an alternative approach for assessing diagnostic coding and to explore the implications of miscoding. The first two sites have clinical data; the second two have administrative data. The prevalences of 14 comorbidities and the associated risk ratios for short-term mortality were compared across data sets. Some comorbidities could not be precisely mapped to ICD-9-CM. Chronic or asymptomatic conditions such as mitral insufficiency, cardiomegaly, previous myocardial infarction, tobacco use, and hyperlipidemia were far less prevalent in administrative data than in clinical data. The prevalence of diabetes, unstable angina, and congestive heart failure were similar in administrative and clinical data. Estimates of relative risk derived from clinical data equalled or surpassed those derived from administrative data for all conditions. Hospitals should be encouraged to improve reporting of coexisting conditions on discharge abstracts and claims. In the meantime, researchers using administrative data should assess the vulnerability of their risk models to bias caused by selective underreporting.

Health care seeking behavior following a health survey: impact on prevalence estimates of chronic diseases.

This article addresses the time sequence between a population health survey and subsequent health care use and how this changes the incidence estimates of selected chronic diseases. A cardiovascular survey of a representative sample of the adult population of Manitoba, Canada was linked with the health insurance claims database. Of the 2792 subjects in the survey, 98% were linked successfully, using an encrypted personal health insurance number. Five years of physician claims data for the survey participants were reviewed including 18 months prior to and 42 months following the survey. Survey participants started seeking confirmation of possible hypertension as soon as they received blood pressure information at the interview. Confirmation of diabetes and elevated cholesterol were not completed until 3-4 months after participants had received the laboratory test results. As many as 4.6 times more new cases of hypertension per month, 5.1 times more cases of elevated cholesterol, and 3.3 times more cases of diabetes were diagnosed following the survey. Surveys designed to determine the prevalence of specific chronic diseases generate new cases within a short time afterwards, thus affecting the original prevalence estimates. The process of assessing the burden of disease in a population is dynamic rather than static, and comparisons across populations need to take into account the frequency and recency of past surveys.

Surveillance of vaccine-related adverse events in the first year of life: a Manitoba cohort study.

This study used data from the population database through which the province of Manitoba, Canada, administers its universal health insurance plan. Enrollment, hospitalization, and immunization files from children born in the 1987-1989 period were linked using the unique identification number assigned to each population member. Analysis of these linked data successfully identified serious potential adverse events in the first year of life and the timing of events around immunization. Not only is population-based active surveillance for immunization-related events feasible, but the techniques described, applied to years of data accumulated through surveillance, offer powerful research tools. Baseline population incidences of adverse events were calculated, temporal relationships between events and immunization assessed, and incidences for events showing true temporal associations determined. Eventual goals are the quantification of vaccine-related risk and the gathering of evidence concerning casual associations. The approach could be used readily by several other Canadian provinces and by health maintenance organizations in the United States.

Changes in asthma severity in Manitoba.

OBJECTIVE: To assess changes in the severity of physician-diagnosed asthma between 1983 and 1988. DESIGN: Cross-sectional studies examining the frequency of markers of asthma severity: hospitalizations, ICU admissions, hospital emergency department visits, multiple physician contacts, and referrals to specialists in patients aged 0 to 14 years, 14 to 34 years, and > or = 35 years separately. SETTING: Physicians' claims data from the universal Provincial Health Insurance Plan for fiscal years 1983 and 1988. PATIENTS: All patients with the diagnosis of asthma, bronchitis, and COPD identified from the Manitoba Health database. MEASUREMENTS: The markers of severity were related to the prevalence of patients seeing a physician and receiving a diagnostic label of asthma, COPD, or bronchitis. RESULTS: The number of patients with physician-diagnosed asthma increased by 36.4% over the 5 years. In 1983, 11% of asthmatics were hospitalized during the year and 8% were hospitalized in 1988 (-2.5%; 95% confidence interval [CI], -3.2 to -1.8%). During both years, about 75% of the patients hospitalized were in hospital once only. Mean and median duration of hospital stay declined. The percentage of asthmatics seen in the hospital emergency departments declined slightly in all age groups, the total being 21% in 1983 and 18% in 1988 (-3.5%; 95% CI, -4.5 to -2.5%). About one third of the patients with asthma were seen only once by a physician during both of the years examined, 43 to 45% of them being seen on three or more occasions during both years. Referrals to specialists for all asthmatics increased from 12 to 14% (1.9%; 95% CI, 1.0 to 2.8%) from 1983 to 1988. This was almost entirely due to an increase from 11 to 16% (5.1%; 95% CI, 4.0 to 6.2%) in the youngest age group, an increase not accompanied by an increase in any other marker of severity. Changes in asthma severity were similar to changes in the severity in patients with bronchitis and COPD. CONCLUSION: No increase in severity of asthma was seen between 1983 and 1988, but the prevalence of the diagnostic label of asthma increased substantially.

Trends in physician-diagnosed asthma prevalence in Manitoba between 1980 and 1990.

We attempted to assess recent changes in the prevalence of physician-diagnosed asthma and the possible influence of diagnostic exchange on these trends. The routinely collected data of the provincial Health Insurance Plan (physicians' claims) were used to determine the annual prevalence of physician-diagnosed asthma in Manitoba. Results indicate that the prevalence of physician-diagnosed asthma increased for all age groups in both male and female subjects between 1980 and 1990. The average increases were the highest in the age group 5 to 14 years for both sexes. The average increases varied with age and there were significant differences between the two sexes. There was evidence of increasing diagnostic exchange, that is, a tendency to label patients with asthma instead of alternative diagnoses. This was particularly prominent in those younger than 35 years of age. However, the increased prevalence of physician-diagnosed asthma, even for the younger population, cannot be fully explained by diagnostic exchange.

Auditing the process of care in a new geriatric unit.

To assess whether the objectives of a new Extended Care Unit were reflected in the care of the patients and in the outcome of that care, an audit of the patients' records was performed. The audit sample involved 101 geriatric patients who had been admitted to the Unit for rehabilitation therapy, and then discharged. The study included assessment of the records for compliance with individual audit criteria, examination of the records in terms of a composite audit score, and analysis of the relationship between these scores and three outcome indices. Although the audit did not show a significant correlation between recorded care processes and treatment outcomes, it did reveal the extent to which the Unit's goals were reflected in the care process.

A framework for modelling differences in regional mortality over time.

STUDY OBJECTIVE: To present a conceptual framework for testing differences in mortality for small geographical areas over time using the generalised linear model with generalised estimating equations. This framework can be used to test whether the magnitude of regional inequalities in health status has changed over time. DESIGN: A Poisson regression model for correlated data is used to investigate the relation of population health status to demographic, geographical, and temporal explanatory variables. Differences between regions at one or more points in time are tested with linear contrasts. SETTING AND PARTICIPANTS: A case example shows the application of the framework. All cause mortality and cause specific mortality were compared for three rural regions of Manitoba, Canada between 1985 and 1999. The data were obtained from Vital Statistics records and the provincial health registry. MAIN RESULTS: Tests of linear contrasts on the regression coefficients for time and region show an increase in the magnitude of the difference in the risk of all cause mortality and heart disease mortality between northern and southern regions of the province for the 1985-1989 and 1995-1999 time periods. No significant differences are identified for cancer, injury, or respiratory disease mortality. CONCLUSIONS: The proposed framework enables testing of a variety of hypotheses about differences between regions and time periods and can be applied to other measures of population health status.

Multiple approaches to assessing the effects of delays for hip fracture patients in the United States and Canada.

OBJECTIVE: To examine the determinants of postsurgery length of stay (LOS) and inpatient mortality in the United States (California and Massachusetts) and Canada (Manitoba and Quebec). DATA SOURCES/STUDY SETTING: Patient discharge abstracts from the Agency for Health Care Policy and Research Nationwide Inpatient Sample and from provincial health ministries. STUDY DESIGN: Descriptive statistics by state or province, pooled competing risks hazards models (which control for censoring of LOS and inpatient mortality data), and instrumental variables (which control for confounding in observational data) were used to analyze the effect of wait time for hip fracture surgery on postsurgery outcomes. DATA EXTRACTIONS: Data were extracted for patients admitted to an acute care hospital with a primary diagnosis of hip fracture who received hip fracture surgery, were admitted from home or the emergency room, were age 45 or older, stayed in the hospital 365 days or less, and were not trauma patients. PRINCIPAL FINDINGS: The descriptive data indicate that wait times for surgery are longer in the two Canadian provinces than in the two U.S. states. Canadians also have longer postsurgery LOS and higher inpatient mortality. Yet the competing risks hazards model indicates that the effect of wait time on postsurgery LOS is small in magnitude. Instrumental variables analysis reveals that wait time for surgery is not a significant predictor of postsurgery length of stay. The hazards model reveals significant differences in mortality across regions. However, both the regressions and the instrumental variables indicate that these differences are not attributable to wait time for surgery. CONCLUSIONS: Statistical models that account for censoring and confounding yield conclusions that differ from those implied by descriptive statistics in administrative data. Longer wait time for hip fracture surgery does not explain the difference in postsurgery outcomes across countries.

Complications, comorbidities, and mortality: improving classification and prediction.

OBJECTIVE: First, to compare the distribution of complications and comorbidities associated with 17 common surgical procedures. We then describe the effect of augmenting an ICD-9-CM version of the Charlson comorbidity index, given the possible confounding of comorbidities and complications, for three common inpatient surgical procedures: coronary artery bypass surgery, pacemaker surgery, and hip fracture repair. DATA SOURCES AND STUDY SETTING: Individuals having one of the above procedures between April 1, 1990 and March 31, 1994, identified from Manitoba Health hospital discharge data, and their extracted records. STUDY DESIGN: Design was cross-sectional and longitudinal using Manitoba data on hospital utilization and mortality. DATA COLLECTION/EXTRACTION: Manitoba hospital discharge abstracts permit identifying whether or not the diagnosis represents an in-hospital complication of care. Two data sets were created for each procedure, one including complication diagnoses and another with complications removed. PRINCIPAL FINDINGS: The degree to which complications contaminated estimation of comorbidity depended both on the procedures studied and on the covariates selected. The unique structure of the algorithm for the Charlson comorbidity index led to complication diagnoses having only a minor effect on the comorbidity score generated. Unless a data set affords the opportunity to remove complication diagnoses, the improvement in comorbidity detection afforded by augmenting the Charlson index, combined with the potential for overestimation of comorbidity, seem sufficiently modest to contraindicate such augmentation.

Across time and space: variations in hospital use during Canadian health reform.

OBJECTIVES: To investigate change in hospital utilization in a population and to discuss analytical strategies using large administrative databases, focusing on variations in rates of different types of hospital utilization by income quintile neighborhoods. DATA SOURCES: Hospital discharge abstracts from Manitoba Health, used to study the changes in utilization rates over eight fiscal years (1989-1996). STUDY DESIGN: We test the hypotheses that health reform has changed utilization rates, that utilization rates differ significantly across income quintiles (defined by the relative affluence of neighborhood of residence), and that these variations have been maintained over time. Our approach uses generalized estimating equations to produce robust and consistent results for studying rates of recurrent and nonrecurrent events longitudinally. DATA EXTRACTION METHODS: Rates of individuals hospitalized, hospital discharges, days of hospitalization, and hospitalization for different types of medical conditions and surgical procedures are generated for the period April 1, 1989 through March 31, 1997 for residents of Winnipeg, Manitoba. Data are grouped according to the individual's age, gender, and neighborhood of residence on April 1 of each of the eight fiscal years for the rate calculations. Neighborhood of residence and the 1991 Canadian Census public use database are used to assign individuals to income quintiles. PRINCIPAL FINDINGS: The substitution of outpatient surgery for inhospital surgery accounted for much of the change in hospital utilization over the 1989-1996 period. Health care reform did not have a significant effect on the utilization gradient already observed across socioeconomic groups. Health reform markedly accelerated declines in in-hospital utilization. CONCLUSIONS: Grouping the data with key characteristics intact facilitates the statistical analysis of utilization measures previously difficult to study. Such analyses of variations across time and space based on parametric models allows adjustment for continuous covariates and is more efficient than the traditional nonparametric approach using standardized rates.

Monitoring adverse outcomes of surgery using administrative data.

In this article, we document a stabilization in adverse outcomes associated with hysterectomies, cholecystectomies, and prostatectomies performed between 1972-73 and 1982-83 in Manitoba, Canada. The proportion of surgery performed by high-volume surgeons and by surgical specialists increased slightly over the decade. However, given the already low rates of adverse outcomes, these changes did not translate into significant decreases in the postoperative mortality rate or in the rate of related hospital readmissions. Reducing the proportion of adverse outcomes would be facilitated by identifying institutions with poorer than expected outcomes.

Issues in studying ancillary services.

This paper examines several approaches to studying ancillary services usage. The likely generalizability of results provides one criterion for picking site, study population and conditions studied. Breaking up such an aggregate figure as mean annual ancillary services charges per patient into its component parts helps identify the different contributors to high cost practice patterns. Potential differences in case mix are noted as providing the most severe threat to interpreting variation in ancillary services use across practitioners and practice settings. Strategies for dealing with case mix problems include: focusing on specific diseases and specific specialities, stratifying within diagnostic categories, using multiple comparisons and multivariate analysis, pairing of visits and episodes and applying small area techniques.