Overview of methods

1
Overview of methods
Paul Fryers Doncaster Public Health Intelligence
Unit
2
Methods for what?

• Health needs assessment
• Community profiling
• Analysis of inequalities in health outcomes and
  access to services
• Setting and monitoring targets
• Evaluation of health services
• Information to support practice-based
  commissioning
• Health impact assessment
• Reporting on public health and providing public
  access to information on health

3
Types of analysis

• Methods of comparison
• Standardisation
• Identifying outliers or significant variations
• Confidence intervals
• Statistical process control
• Identifying and studying relationships
• Regression correlation
• Time series/monitoring trends targets
• Regression
• Exponential smoothing/Forecasting
• Spatial analysis

4
Particular analytical issues within PCTs

• Populations
• Almost everything we do depends on population
  data and if they are wrong so are all the results
  
• Repetition
• We have to do these calculations a lot
• Very boring, and very likely to make errors
• Therefore we have to automate where possible
• Presentation
• We produce large volumes of information for a
  wide range of audiences
• Therefore we have to make the data clear,
  understandable and (preferably) pretty

5
Tools available 1

• Access and other databases
• Hold large datasets
• Allow rapid number-crunching/preliminary analysis
• Excel
• Can handle most data analysis and presentation
• Very visual easy to check through analysis
• Visual Basic saves eons in repetitive processing
  and turns Excel into a software development tool
• Specialist software
• GIS MapInfo most common in PCTs
• Stats packages SPSS StatsDirect ForecastPro

6
Tools available 2

• Statistical Software
• General packages like SPSS sometimes available
• StatsDirect is specifically for Public Health, is
  cheap and bolts on to Excel
• ForecastPro helps with forecasting/time series
  analysis
• Geographical Information Systems (GIS)
• MapInfo most common in PCTs
• Web Sites
• Used for publishing/disseminating information
• Instant Atlas, etc.

7
Support for this analysis

• Analytical support for public health varies
  hugely across PCTs
• All PCTs should either
• have public health intelligence teams who can do
  most of this analysis
• or
• have people who are aware of what is needed and
  can call on support from networks/PHO

8
Standardisation 1

• If you are trying to compare two populations, be
  they districts, neighbourhoods or practices, most
  things need to be standardised for age and sex
  before you can start to make any sense of them
• Deaths
• Hospital referrals/ admissions
• Cancer incidence
• Prevalence of CHD, diabetes, obesity, smoking,
  hypertension, etc. etc. etc.
• You might also want to standardise data for other
  factors, e.g. ethnicity, deprivation, distance
  from hospital

9
Standardisation 2

• Standardisation can be done in different ways
• Direct standardisation is appropriate for
  comparing different populations, all standardised
  to a common reference population or producing
  league tables
• Indirect standardisation is only appropriate for
  comparing each population with the standard
  reference population, not with each other
• Each has advantages for presentation in different
  circumstances examples will be seen in later
  presentations

10
Confidence intervals

• Different methods are required for calculating
  confidence intervals around
• Counts (or observed numbers for indirectly
  standardised ratios, or expected numbers in the
  case of statistical process control/funnel plots)
• Proportions (percentages)
• Directly standardised rates
• Odds ratios
• Sometimes, approximate methods (such as the
  normal approximation) can be used but we are
  often dealing with very small numbers

11
Example Funnel Plot
12
Example Directly standardised rates with 95
confidence intervals
13
Example adjusting for deprivation

• Estimated prevalence of CHD, before and after
  taking account of deprivation

Each line on the graph represents one general
practice
14
Identifying and studying relationships

• First thing is usually to plot two variables
  against each other
• Correlation is often visible
• Correlation coefficient gives indication of
  closeness of relationship
• Correlation is not appropriate for measuring
  agreement between two measures of the same thing
  (see Bland JM, Altman DG. Statistical methods for
  assessing agreement between two methods of
  clinical measurement. Lancet 1986i307-10.)

15
Example relationship between health, gender and
deprivation

• Life expectancy across wards in South Yorkshire

Life expectancy based on deaths between 1995 and
1999
16
Example relationship between need and resource

• 04 year olds per WTE health visitor and
  deprivation
• Implications for resource allocation

17
Thank you
www.doncasterhealth.co.uk/phiu