Download presentation
Presentation is loading. Please wait.
Published byJosiah Anne Modified over 10 years ago
1
Manitoba Centre for Health Policy The Future of Information-Rich Environments Leslie L. Roos Lisa Lix CPS Meetings – June 2004
2
Entry Points 1)Use of Administrative Data from Defined Populations Small area analysis and population-based studies. 2)Development of Record Linkage Individual-based Permits: a.Building histories for each individual to track utilization before or after an index event b.Maintenance of registries c.Management of multifile databases d.Working across data sets
3
Figure 1. Administrative data in an information-rich environment
4
Producing Integrated Systems 1)The presence or generation of a common identifier across data files 2)Longitudinal data files with considerable amounts of individually-based information 3)A population registry with geographic codes 4)Tools to access them
5
Table 1. Manitoba Research Registry: Characteristics and Relevance CharacteristicsResearch Relevance Very Large NMany physical and statistical controls feasible Population-based for an entire province Heterogeneity on many variables Longitudinal data (going back over 30 years) Many types of longitudinal studies, drawing of cohorts, more reliable measurement of important variables
6
Table 1. Continued Specification of place of residence (by postal code) at any time point Neighborhood longitudinal studies; Permits analysis of small area variation Mobility/migration and loss to follow-up well specified Cohort studies; Mobility data allow capturing “length of exposure” Family and sibling information (ties in with neighborhood information) Non-experimental designs estimate importance of different factors control for unobserved / unmeasured background characteristics
7
Table 2. New Data Sets Data SetResearch Relevance Education Grade 3 enrollment and examination results for one year Studies based on birth cohorts can look not only at “pass/fail” outcomes but at school enrollment (grade level or not), non-enrollment but residence in province, and loss to follow- up. Grade 12 enrollment, examination results, and graduation for seven years
8
Table 2. Continued Family Services Received “income assistance” at any time in a seven year period: month by month data Income assistance can be a dependent or independent variable Neighborhood Information Compositional: aggregated upwards from Canadian Census Better understanding of neighborhoods; Characteristics can be entered as independent variables in analyses of health and health care Contextual – collected from various sources (statistics on crime, social programs, etc.)
9
Table 3. Social Variables as Independent Predictors of Human Development Social VariablesData Sets Number of years received family assistance FA Average household income (number of years recorded, partial coverage) P Average household income (from Census enumeration / dissemination area) CS Number of children in the familyRR Data sets: FA (Family Assistance), P (Pharmaceutical), RR (Research Registry), H (Hospital Abstracts), PC (Physician Claims), CS (Canadian Census)
10
Table 3. Continued Mother’s marital status at birth of first childRR Number of household location movesRR Number of years living in a single-parent family RR Age of mother at birth of first childRR Number of “family structure” changes (parental separations, remarriages) RR Data sets: FA (Family Assistance), P (Pharmaceutical), RR (Research Registry), H (Hospital Abstracts), PC (Physician Claims), CS (Canadian Census)
11
Table 3. Continued Individual was first-born childRR Number of years living with a disabled parent H, PC, RR Neighborhood characteristics number of years living in neighborhood with “bad” characteristics RR, CS Data sets: FA (Family Assistance), P (Pharmaceutical), RR (Research Registry), H (Hospital Abstracts), PC (Physician Claims), CS (Canadian Census)
12
Table 4. Quality of Diagnostic Information from Hospital Abstracts and Physician Claims DiagnosisDetailed chart review 1 Abstracts / claims compared with clinical information Abstracts / claims compared with population- based survey Prevalence estimate 5 (survey as base) Asthma (Adult).60 2 Recent Acute Myocardial Infarction 0.750.82 4 Diabetes0.740.72 5 +5% Chronic Pulmonary Disease 0.72 Hypertension0.60 3 0.59 5 -19% Congestive Heart Failure 0.800.58 4 Liver Disease0.75
13
1 Alberta Chart Review (Quan et al., 2002). 2 Canadian MultiCentre Asthma Study (Manitoba Component) (Huzel et al., 2002). 3 Manitoba Heart Health Project (Muhajarine et al., 1997). 4 Ontario Fastrak 11 Acute Coronary Syndromes Registry (Austin et al., 2002). 5 Manitoba Heart Health Project (Robinson et al., 1997). Table 4. Notes
14
Statistics for Information- Rich Environments Random effects models Models for multiple outcome variables Spatial regression models
15
Random Effects Models Regression coefficients can vary across subjects Components 1)Within-individual component –Individual’s change over time is described by a regression model with a population-level intercept and slope 2)Between-individual component –Variation in individual intercepts and slopes is captured
16
Advantages of Random Effects Models Ability to incorporate time-varying covariates –Severity of illness –Presence of co-morbid conditions –Location of residence Development of non-linear models –Count data –Binary data
17
Random Effects Models: Example Health service use and cost at the end of life Measurements compiled each month of the last six months of life Partition between-individual and within-individual variation –Between-individual variation in trends over time accounts for a substantial portion of variation in the data Explanatory variables –Age, gender, income quintile, region of residence, location of death, cause of death
18
Spatial Regression Models Account for spatial auto-correlation in the data, which can lead to biased regression parameter estimates Example: Geographically weighted regression (GWR) –Individual parameter estimates are produced for each geographical area, along with the associated standard errors, test statistics, and p-values
19
Models for Multiple Outcome Variables Simultaneous models for two or more outcome variables Controls Type 1 errors arising from multiple hypothesis testing on individual dependent variables Accounts for correlation between outcomes, which may increase statistical power
Similar presentations
© 2024 SlidePlayer.com. Inc.
All rights reserved.