Health measurement in population surveys: Combining information from self reported and observer meas

1
Health measurement in population surveys
Combining information from self reported and
observer measured health indicatorsGeorge B.
Ploubidis
2

• In the quest for true population health a
  plethora of health indicators have been used
• Broadly they are classified into self reported
  and observer measured performance based-
  indicators
• Both types of health indicators are susceptible
  to measurement error

3
Self reported health indicators

• Self rated health (the most widely used)
• Self report of chronic illness
• Self report of somatic symptoms
• Activities of Daily Living/Instrumental
  activities of daily living

4

• Self rated health is probably the most widely
  used indicator of true population health status
• Easy to use and is the strongest predictor of
  all cause mortality
• But!!! Problems with reporting styles (applies
  to all self reported indicators)
• Measurement invariance between SEP groups also a
  problem
• But, what does it really measure? (individual
  differences in health attribution as well as
  prior information)

5
Observer measured indicators

• Grip strength
• Spirometry (Forced Vital Capacity, Forced
  Expiratory Volume and others)
• Blood pressure
• Timed chair rise stands
• Blood analytes
• Missing data can be a problem, especially in
  the older population, where the most frail have a
  higher probability of not participating in the
  measurements

6
But what happens when true population health is
the outcome of interest?

• Barsky paradox Self reported and observer
  measured health indicators often in conflict -
  especially in estimation of health trends over
  time
• A more reliable estimate of true population
  health status is needed

7
Why Latent Variable Models?

• LVM provide an opportunity to combine self
  reported and observer measured indicators to a
  reliable estimate of population health
• According to psychometric theory responses to a
  health indicator can be decomposed to true health
  and measurement error
• Measurement error can be further decomposed to
  systematic and random
• Big (but testable) Assumption!!!!!!!!!!!!!
• There is a latent dimension common to all
  these indicators that represents health

8
Why Mplus?

• Only software that can combine binary, ordinal
  and continuous indicators in a latent variable
  within a single measurement model
• Measurement model can be used as outcome or
  predictor with various models, e.g Survival
  models, Growth Curve models, LCA models and more

9

• Method
• We used data from a nationally representative
  sample of the older population, the 2nd wave of
  the English Longitudinal Study of Ageing (ELSA)
  which was carried out in 2004
• Six health indicators were used, three self
  reported and three observer measured
• Participants were included in the analysis if
  they had data on all six indicators, N 5,964
  (out of 8,780 total) We also estimated models
  with missing data

10
Descriptive statistics of the 6 health indicators
11
A unidimensional measurement model
Grip strength
e
Chair rise
e
Lung function
Somatic Health
e
Functional limitations
e
Self rated health
e
Long standing illness
e
12
A 2nd order model
Grip strength
e
Observer measured
Chair rise
e
Lung function
Somatic Health
e
Functional limitations
e
Self reported
Self rated health
e
Long standing illness
e
13
A bifactor multimethod model
Grip strength
e
Observer measured
Chair rise
e
Lung function
e
Somatic health
Functional limitations
e
Self reported
Self rated health
e
Long standing illness
e
14

• We carried out analyses with MLR and WLSMV Very
  similar results
• The selected model was regressed on covariates
  (MIMIC model) that represent SEP as well as
  external validation criteria

15

• Results
• The bifactor model was confirmed by the data and
  was superior compared to the two competing
  measurement models

16
Standardised factor loadings and proportions of
the health indicators explained by the model

• Self report of Functional Limitations is the most
  reliable health indicator!!!!

17
The distribution of somatic health in the
population
High values indicate good somatic health
18
MIMIC model standardised parameters
19
External criterion validity I
20
External criterion validity II
21
External criterion validity III
22

• Conclusions I
• Our results support a single latent dimension,
  common to both self reported and observer
  measured health indicators ,that represents
  somatic health
• This latent dimension is a reliable quantitative
  estimate of true population (somatic) health
• Measurement invariance of this population health
  metric over groups or time can me easily tested
• Response bias and other forms of measurement
  error are largely controlled

23
Conclusions II

• Self reported and observer measured health
  indicators appear to be equally reliable, thus
  both need to be considered in population health
  measurement
• FL is the indicator of choice in the older
  population when time/budget constrains allow for
  one health indicator to be used
• Responses of both genders to self rated health
  items are biased by exogenous to their somatic
  health status influences

24
Future plans

• Add mental health to the model as an indicator of
  health
• OR joint modelling of somatic and mental health
  outcomes (probably better)
• Apply the model to the younger population and use
  different health indicators
• Use the LISH in the calculation of Healthy life
  Expectancy currently done with self rated
  health and functional limitations only

25

• Limitations
• The use of chair rise as an indicator of
  health excluded the most frail participants from
  the analysis, although models including missing
  data returned similar results
• The indicators employed in this study are a
  sample of the population of possible health
  indicators.