Transcript Ontario Cancer Risk Factor Surveillance Program

The Ontario Cancer Risk Factor
Surveillance Program
Michael Spinks
Senior Research Analyst
Cancer Care Ontario
at
5th Annual RRFSS Workshop
Institute for Social Research, York University
June, 2006
Contents






Risk Factor Surveillance at CCO
CCO analysis of RRFSS data
Generating complex survey estimates using SPSS
Risk factor indicator inference and trends
CCO Risk Factor Surveillance Reporting System
Next Steps
CCO Cancer Risk Factor Surveillance
System

CCO is very supportive of RRFSS

Risk Factor Surveillance Project established at CCO

Important to liaise with suppliers and users of risk
factor data
Risk Factor Surveillance Methodology
Data Sources

RRFSS (monthly survey, available in 6 weeks)

CCHS (annual survey, available in 6 months)

Other Survey and Related Data (OHS, NPHS, OBSP,
SHAPES, OSDUS)

Population Estimates and Projections

Census data
Risk Factor Surveillance Methodology
Indicator Development




Cancer 2020 project
Review of indicator definitions from other agencies –
CWIG(APHEO), RRFSS, Statcan, camh
Develop indicators using flow diagrams and existing
survey data
Indicator refinement and standardization
(Beth Theis – CCO representative on CWIG)
Current Risk Factor Indicators by Survey
Indicator
CCHS
RRFSS
Adult smoking


Teen smoking

Quitting smoking


Exposure to 2nd hand smoke


Adult obesity


Teen obesity

Physical activity


Alcohol consumption


Fruit & Vegetable Intake


Mammography screening


Cervical screening


Colorectal screening

Sun safety

Tanning equipment usage

Risk Factor Surveillance Methodology
Survey Analysis Review


Single-stage sampling
- random selection of individuals from the population is sampled
- for a simple random sample, each sample of a given size is equally
likely to be selected from the population
- each individual has the same probability of being selected
- computation of point and variance estimates relatively straightforward
Multistage sampling
- units at the first stage are clusters of individuals
(or clusters of smaller clusters)
- mainly used for cost and logistical reasons
- individuals have unequal probabilities of being selected
- variability or estimates greater compared with simple random sample
of same size
- computations of point and variance estimates more complex
Risk Factor Surveillance Methodology
RRFSS Survey Design
At provincial level RRFSS considered to be a multistage cluster sample design
stage 1 cluster (PHU)
and stage 2 cluster (household)
PHU and CCO Weighting Procedures



What is the sampling weight
- each individual represents other persons not in sample
- computed as the inverse of the inclusion probability
- used to obtain unbiased estimates of risk factor indicators
Sample weight used by PHU (monthly/annual)
- inclusion probability of selecting an adult member from sample of
households
- weights total to number of respondents in sample
Sample weight used by CCO (annual)
- inclusion probability of selecting an adult member in the population
- adjusted so each month is equally represented
- adjusted to size of population age/sex structure
- weights total to number of adults in population
Respondents by PHU and Wave, 2004
Number of respondents vary slightly by month
Comparison of Estimates – PHU and CCO


Point estimates
- Both methods yield almost identical point estimates
Variance estimates
- Assuming simple random sampling (PHU)
- Taylor’s series linearization (CCO)
- Bootstrap resampling (CCHS)
- Jack-knife resampling
- Balanced half-sample
Comparison of estimates - PHU and CCO Approaches
PHU approach underestimates variance of multistage survey design
Comparison of estimates - PHU and CCO Approaches
Was the percentage
of smokers in
Durham significantly
lower in 2003 than in
2001?
Tools for computing estimates from
complex surveys

SAS (CCO) – proc surveyfreq, surveymeans, surveyreg,
surveylogistic



SPSS (PHU) - CSPlan then - CSDescriptives, CSTables,
CSTabulate, CSGLM, CSLogistic
Sudaan – proc crosstab, descript, ratio, regress, logistic
Stata – svyset, then svy: mean, proportion, ratio, total, regress,
logit, etc.
1
Computing estimates from complex surveys in SPSS
3
SPSS Syntax
2
* Analysis Preparation Wizard.
CSPLAN ANALYSIS
/PLAN FILE='M:\RRFSS\SPSS\rrfssplan.csaplan'
/PLANVARS ANALYSISWEIGHT=fwgt
/PRINT PLAN
/DESIGN STRATA= h_unit CLUSTER= idnum
/ESTIMATOR TYPE=WR.
Computing estimates from
complex surveys in SPSS
Computing estimates
from complex surveys
in SPSS - Results
Comparison of estimates generated from SPSS and SAS
% of current smokers, Durham Regional Health Unit, 2004
Sex
Age group
SPSS
%
male
female
SAS
se
%
se
18-44
34.01798
3.080043
34.01798
3.080045
45-64
24.99574
3.380777
24.99574
3.380780
65+
15.16691
4.413263
15.16691
4.413266
18+
28.79788
2.123433
28.79788
2.123434
18-44
28.44647
2.656429
28.44647
2.656430
45-64
21.90933
2.851722
21.90933
2.851724
65+
9.18770
2.939705
9.18770
2.939707
18+
23.49117
1.759892
23.49117
1.759893
• Estimate of point statistic identical
• Estimate of standard error identical to the 5th decimal place
CCO Risk Factor Measures
Compute range of statistics for different indicators to be able to respond
to the majority of analytical needs
Indicator definitions
Point statistics
Statistics for evaluating precision
Multiple combinations of
numerators and
denominators as required
e.g.
for female low alcohol
risk
1. <=9 drinks/week
2. <=9 drinks/week and
<=2 drinks daily in last
week
Counts/Prevalence ratios
Sex and/or age specific
Crude/Age standardized
PHU/LHIN/Province
95% confidence intervals
Standard errors
Coefficient of Variation (CV)
Numerator/Denominator sample sizes
Risk Factor Estimates at the
Provincial Level




Almost 100% of population and 100% of Health Units
represented in CCHS
85% of population and 67% (24) Public Health Units
represented in RRFSS 2004
Estimates from RRFSS Public Health Units are not usually used
as a proxy for the province
RRFSS not representative of northern PHUs
Comparison of Risk Factor Estimates between
RRFSS Health Units and Non-RRFSS Health Units
using CCHS 2.1
Prevalence of Selected Risk Factor Indicators with 95% CI
All RRFSS Health Units
All Non-RRFSS Health Units
60.0
50.0
40.0
% 30.0
20.0
•Significantly different at 5%
Data Source: CCHS 2.1, Statistics Canada
male
f emale
physically
active *
fruit/vegetable
(5+/day)
obese *
current
smokers *
physically
active
fruit/vegetable
(5+/day) *
obese *
0.0
current
smokers *
10.0
Comparison of Risk Factor Estimates

Overlapping confidence intervals

Compute age-standardized rates (age groups-12-17, 18-44, 45-64, 65+)

Funnel plots for comparing PHUs

Significance testing using logistic regression and controlling for age
and sex differences
Risk Factor Surveillance Methodology
Trends

Annual plots of RRFSS and CCHS estimates

Quarterly plots of RRFSS estimates

Change point analysis

Control charts

Box-jenkins time series analysis
PrevCan CCO Risk
Factor
Surveillance
Reporting
System
Next Steps






Collaboration with CE RRFSS Group
Establish agreement with RRFSS for sharing of data and
technical support
Share developments with MOHLTC
Refine methods for testing and dissemination of results
Expand indicators to include socio-economic and environmental
factors
Include GIS in risk factor surveillance