Healthy Worker Effect (HWE): William Ogle, 1885

1
Healthy Worker Effect (HWE)William Ogle, 1885

• Identified two major difficulties in occupational
  mortality studies
• some occupations may be of necessity recruited
  from men of supernormal physical condition
  (selection into the workplace or cohort)
• various occupations require a considerable
  standard of muscular strength and vigor to be
  maintained (survivor effect, or selection out of
  the workplace or cohort)

2
Healthy Worker Effect (HWE)William Ogle, 1885

• Ogle recognized that the bias from these two
  difficulties was in the same direction bias away
  from the null and towards lower than expected
  mortality and morbidity rates

3
Healthy Worker Effect (HWE)

• A pervasive problem in occupational mortality and
  morbidity studies
• Leads to underestimates of, (or may totally
  obliterate), exposure-disease associations
• Leads to distortion of dose-response
  relationships
• General principle for minimizing HWE bias
• make exposure-disease comparisons among those
  similarly affected by the bias

4
Sources of Healthy Worker Effect

• Healthy Hire Effect
• Initial selection of healthy individuals at time
  of hire so that their disease risks differ from
  the disease risks in the source (general)
  population
• Hiring is dependent upon passing a pre-employment
  medical examination
• Self-selection of individual (able to apply for
  the job)
• Bias occurs when comparing disease rates between
  the worker cohort and the general population
• Effect is greatest during the initial period of
  follow-up and diminishes to some extent over time
• Bias lower than expected mortality/morbidity
  rates

5
Sources of Healthy Worker Effect

• Time-Since-Hire Effect
• Decline in health with time (e.g., gt15 years)
  since hire (the effect of the initial selection
  in the hiring process wears off over time)
• Bias occurs because lower cumulative exposure
  categories include more recent hires than higher
  cumulative exposure categories
• Increasing time-since-hire is correlated with
  increasing cumulative exposure
• Higher cumulative exposure categories appear to
  be associated with higher disease risks even when
  no exposure-disease relationship exists
• Bias is away from null leading to an overestimate
  of the health effect at higher cumulative
  exposure levels

6
Sources of Healthy Worker Effect

• Healthy Survivor Effect
• a continuing selection process in the workplace
• Survival of healthier individuals in the active
  workforce (a selective retention of healthy
  workers in the workforce)
• Selection out of unhealthy (symptomatic) workers
  from workforce (health-related job mobility or
  termination)
• Terminate employment
• Transfer to jobs with lower exposures

7
Sources of Healthy Worker Effect

• Healthy Survivor Effect
• Can be viewed as a confounder
• Terminating employment (or transfer to a less
  exposed job) is related to future exposure
• Terminating employment is an independent risk
  factor for disease
• Beneficial effects to health of continued
  employment
• Improved access to health care
• Higher standard of living
• Physical exercise
• Routine disease screening at workplace
• Job transfer may be related to health and
  exposure status
• Workers realize exposure is causing health
  problems and transfer to lower/no exposure jobs

8
Sources of Healthy Worker Effect

• Healthy Survivor Effect
• Can also be viewed as a selection bias
• Impact on a cross-sectional study of a cohort
• Study includes only workers remaining in the
  workplace (active workers) a survivor
  (healthier) population
• Less healthy workers leave (i.e., are selected
  out of) the workplace prior to the study and are
  therefore not included in the study
• Bias effects of exposure are underestimated
• Impact on longitudinal (prospective/retrospective)
  cohort study
• Healthier workers remain in the workplace and
  therefore generally have higher cumulative
  exposures than less healthy workers who leave the
  workplace or transfer to less exposed jobs
• Effects of increasing cumulative exposure are
  underestimated
• The bias is not due to a loss to follow-up, since
  those who leave work are not usually lost to
  follow-up.

9
Sources of Healthy Worker Effect

• Healthy Survivor Effect Bias
• Leads to lower than expected disease rates
• When there is no exposure-disease relationship,
  (and even after stratifying by time-since-hire),
  higher cumulative exposure appears protective of
  health
• Workers with longer employment history will, on
  average, have higher cumulative exposures than
  workers with shorter employment history, and
• Since workers with shorter employment history
  tend to be less healthy due to the healthy
  survivor effect, higher cumulative exposures can
  appear to have a protective effect

10
Sources of Healthy Worker Effect

• Some industries may accept workers with social
  problems, unhealthy habits, or health problems
• contributes to healthy survivor effect bias
• Some industries will not hire such workers
• contributes to healthy hire effect bias
• Some factors affecting hiring practices
• Unemployment rate
• Prevalence of pre-employment health exams
• Availability of other social safety net
  programs
• Physical requirements of the job
• If selection into subgroups of job tasks or
  exposures within the cohort is based on health
  status of the worker at time of hire, then
  internal comparisons may be biased (internal
  healthy hire effect bias)

11
Sources of Healthy Worker Effect

• The effects of healthy worker effect biases may
  vary by
• Gender
• Race/ethnicity
• Social class
• Work status
• Age at hire
• Length of employment
• Length of follow-up
• Type of occupation
• Cause of mortality/morbidity

12
Healthy Worker EffectImpact on measures of
effect

• Impact (bias) on effect measures is most striking
  for
• morbidity measures
• pulmonary function, respiratory symptoms
• neurobehavioral symptoms
• workplace injuries
• mortality measures
• all causes of death combined
• nonmalignant chronic conditions (e.g., heart
  disease)
• Impact is less striking for cancer endpoints

13
Retrospective Cohort Study of polyvinyl chloride
manufacturing workers, 1940-1974Healthy Hire
Effect
Standardized Mortality Ratio (SMR) by cause of death and length of time (years) since entering the industry Standardized Mortality Ratio (SMR) by cause of death and length of time (years) since entering the industry Standardized Mortality Ratio (SMR) by cause of death and length of time (years) since entering the industry Standardized Mortality Ratio (SMR) by cause of death and length of time (years) since entering the industry Standardized Mortality Ratio (SMR) by cause of death and length of time (years) since entering the industry Standardized Mortality Ratio (SMR) by cause of death and length of time (years) since entering the industry
Cause of Death 0-4 yrs 5-9 yrs 10-14 yrs 15 yrs Total period
All causes 37.4 62.9 75.1 94.2 75.4
All cancers 44.5 70.6 94.0 118.8 90.7
Circulatory disease 21.5 70.3 84.7 90.7 76.9
Respiratory disease 20.9 38.8 31.3 93.0 62.6
14
Retrospective Cohort Study of polyvinyl chloride
manufacturing workers, 1940-1974Healthy
Survivor Effect
Current and past (i.e., terminated employment lt15 yrs from time of hire) employees alive 15 years after time of hire Current and past (i.e., terminated employment lt15 yrs from time of hire) employees alive 15 years after time of hire Current and past (i.e., terminated employment lt15 yrs from time of hire) employees alive 15 years after time of hire Current and past (i.e., terminated employment lt15 yrs from time of hire) employees alive 15 years after time of hire Current and past (i.e., terminated employment lt15 yrs from time of hire) employees alive 15 years after time of hire Current and past (i.e., terminated employment lt15 yrs from time of hire) employees alive 15 years after time of hire Current and past (i.e., terminated employment lt15 yrs from time of hire) employees alive 15 years after time of hire
Cause of death Current employees Observed Expected SMR Current employees Observed Expected SMR Current employees Observed Expected SMR Past employees Observed Expected SMR Past employees Observed Expected SMR Past employees Observed Expected SMR
All cancers 24 26.99 89 44 33.81 130
Lung cancer 6 11.91 50 22 14.10 156
Circulatory disease 37 49.24 75 73 72.01 101
Respiratory disease 8 12.81 63 24 21.63 111
All deaths 75 101.36 74.0 155 142.94 108.4
15
Healthy Worker Effect (HWE)Impact on measures
of effect

• Why is HWE bias greater for nonmalignant
  morbidity/mortality?
• Symptoms usually accompany these conditions
• Asymptomatic individuals are
• more likely to be hired
• more likely to remain actively employed
• less likely to either leave the workplace or
  transfer to a job with lower exposures

16
Minimizing Healthy Worker Effect Biases

• Minimizing Healthy Hire Effect Bias
• Make comparisons internal to the cohort
• e.g., compare disease rates among different
  exposure categories within the cohort
• Internal comparisons minimize this bias because
• All cohort members passed through similar hiring
  process
• Differences in the distribution of confounders
  (e.g., smoking status) within the cohort are much
  smaller than differences between the cohort and
  the general (source) population
• Avoid combining all causes of mortality

17
Minimizing Healthy Worker Effect Biases

• Minimizing Time-Since-Hire Effect Bias
• Internal comparisons alone cannot solve the
  problem
• Stratify by time-since-hire
• Compare cumulative exposure groups (e.g., high vs
  low cumulative exposure) among those with shorter
  time since hire
• Compare cumulative exposure groups among those
  with longer time since hire
• Stratify by age (currrent or at hire) and work
  status (active/inactive) instead of
  time-since-hire

18
Minimizing Healthy Worker Effect Biases

• Minimizing Healthy Survivor Effect Bias
• The most difficult of the healthy worker effect
  biases to minimize
• Making internal comparisons and stratifying by
  time-since-hire are not sufficient to minimize
  this bias
• General principle make comparisons among those
  who are similarly affected by the healthy
  survivor effect bias

19
Possible strategies to Minimize Healthy Survivor
Effect Bias

• Restrict analysis to those who survive and are
  followed up for at least 10-15 years since time
  of hire
• Disadvantages
• loss of statistical power due to smaller numbers
  after restriction
• Assumes healthy survivor effect is minimal after
  10-15 years, but there is no evidence that this
  is the case.

20
Possible strategies to Minimize Healthy Survivor
Effect Bias

• Stratify by employment status active vs inactive
• Disadvantage
• Inactive group has heterogeneous disease risks,
  since the inactive worker may be
• employed elsewhere
• disabled
• retired
• voluntarily unemployed or involuntarily
  unemployed
• Solution
• obtain data to distinguish those who are
  off-work from those who are employed elsewhere
• If data are unavailable, stratify by current age
  (a proxy for retired vs other inactive)
• Stratify by time since transfer to deal with
  workers who transferred for health reasons

21
Possible strategies to Minimize Healthy Survivor
Effect Bias

• Stratify by employment status active vs inactive
• Disadvantage
• Cross-sectional studies only include active
  workers, and there is no data on the health
  status of inactive workers
• Solution
• Transform prevalence data into incidence data
• use self-reported year of first onset of symptoms
  to determine yearly incidence rates
• compare incidence rates among the exposure groups
  for each year prior to the cross-sectional survey
• Focus on incidence rates one to two years prior
  to the survey date

22
Example of transforming prevalence data into
incidence data by using self-reported year of
onset to determine yearly incidence rates
persistent pain from repetitive motion among
garment workers
23
Possible strategies to Minimize Healthy Survivor
Effect Bias

• Lag exposures ignores recent exposures
• Motivation Since only the healthier workers
  survived on the job to receive recent exposures,
  ignore the recent exposures to eliminate the
  relationship between exposure and job survival.
• This is the same procedure as assuming a latency
  period for a disease, except that the motivation
  for defining a latency period has to do with the
  exposure-disease process, not the exposure-job
  survival relationship

24
Possible strategies to Minimize Healthy Survivor
Effect Bias

• Lag exposures
• Disadvantages
• Assumes that the period in which the healthy
  survivor bias operates is shorter than the
  exposure lag time (or latency period)
• Assumes time off work is equivalent to time on
  work at zero exposure
• Empirical results indicate that this approach
  works as well or better than stratifying on
  employment status in dealing with bias due to
  termination of employment for health reasons