The Impact of Workplace Policies and Other Social Factors on Self-Reported Influenza-Like Illness Incidence During the 2009 H1N1 Pandemic

SOCIAL DETERMINANTS OF POTENTIAL EXPOSURE TO INFLUENZA VIRUS

In 2008 Blumenshine et al.⁹ proposed a model predicting that unequal levels of illness and death in a pandemic would be affected by social determinants, including household crowding, inability to take time off work, and dependence on public transportation. They hypothesized that higher levels of crowding and higher prevalences of certain types of occupation among certain population groups may give rise to disparities in exposure. Difficulty avoiding public transportation would be another source of disparity in exposure to infectious agents. Finally, staying away from work, if used as a social-distancing policy during a pandemic, would likely be more difficult for lower-wage workers because they would be less able to afford the subsequent loss of income.⁹

Researchers have speculated that disparities in hospitalization and mortality in the 2009 H1N1 pandemic may have been attributable to unequal exposure.³ Those who live in a metropolitan area¹⁰ or in crowded locales, such as an apartment building, may have had higher levels of exposure and higher influenza incidence rates. Furthermore, household crowding may also have affected ILI incidence.^9,11

Workplace policies could affect differential exposure to virus and disease incidence. During the early part of the 2009 H1N1 pandemic, the Centers for Disease Control and Prevention (CDC) recommended that those with ILI stay home from work for 7 to 10 days; as epidemiological information on the illness became available, the CDC issued an updated recommendation for sick people to stay home for an additional 24 hours after symptoms subsided.^12,13 In the United States, the Bureau of Labor Statistics reports that 33% of the civilian workforce lacks paid sick leave.¹⁴ Those who cannot take time off from work, who are unable to work from home, or who lack sick leave at work are at higher risk for exposure via colleagues not staying home when ill.¹⁵ Therefore, these policies may predict ILI incidence. Such worksite policies also hamper workers' ability to rest and recuperate after disease has developed.^16,17 In the first study to operationalize constructs from the Blumenshine et al. model, Quinn et al. showed that multiple factors might influence racial/ethnic disparities in the pandemic in the United States and that risk of exposure to H1N1 was significantly related to race and ethnicity.¹¹ However, the factors that actually affect disease incidence have not been identified.

Dependence on often-crowded public transportation may be associated with increased exposure to influenza and ILI incidence. In the United States, more than 59% of public transportation users are racial/ethnic minorities.¹⁸

Objective data on hospitalization as a result of H1N1 infection during the 2009 pandemic are available, but incidence rates have had to be estimated by using a variety of data sources, including self-reported ILI.¹⁹ The CDC uses the Behavioral Risk Factor Surveillance System to collect self-reported ILI data.^20,21 In this study, we collected data on self-reported ILI incidence during the 2009 H1N1 pandemic, and we examined the data as an outcome of social determinants of potential exposure to the virus. To empirically test the Blumenshine et al. conceptual model,⁹ we developed a path analysis for ILI incidence predicted by social determinants of potential exposure to the virus and social position.

METHODS

We surveyed a nationally representative random sample of adults from the Knowledge Networks online research panel²² about their social and demographic characteristics that were relevant to the H1N1 pandemic. Knowledge Networks uses a combination of random-digit dial and address-based probability sampling to recruit panelists and provides access to computer equipment and Internet service if needed. The panel is thus designed to be representative of the entire US population, including unlisted, nontelephone, and cell phone–only households.

For this study, a national sample of 3689 adults aged 18 years or older, including oversamples of African Americans and Hispanics, was contacted by e-mail. Between January 22 and February 1, 2010, 2079 respondents completed the survey, for a completion rate of 56%. Knowledge Networks provided a data file with weighting variables, which incorporated design-based weights to account for recruitment of the panelists and both panel-based and study-specific post-stratification weights benchmarked against the most recent Current Population Survey (CPS) with respect to demographic and geographic distributions of the US population aged 18 years and older. We analyzed responses from 2042 respondents who reported their race/ethnicity as White non-Hispanic, Black non-Hispanic, or Hispanic (note that we will use “Black” to denote non-Hispanic Black and “White” to denote non-Hispanic White). We excluded the “Other” race/ethnicity category because its small size and heterogeneity made interpretation of the results by race/ethnicity difficult.

RESULTS

Table 1 shows demographic characteristics and ILI incidence for the sample. There were significant differences by race/ethnicity in age, income, education, and household size (P < .001). Ten percent of the respondents reported having had ILI during the pandemic. Thirteen percent reported that someone in their household had had ILI during the pandemic, and this result differed by race/ethnicity (P < .001).

DISCUSSION

We have shown that social factors, including work-related inability engage in social distancing and household crowding, were related to a higher self-reported ILI incidence in the 2009 H1N1 pandemic. In addition, these social factors were more prevalent among Hispanics, putting them at greater risk for ILI.

Quinn et al. used data collected during the first wave of the H1N1 pandemic and showed that Hispanics who took their survey in Spanish were at greater risk than were Whites of potential exposure attributable to worksite policies and household size.²³ Blake et al. failed to detect racial/ethnic disparities in the ability to get paid if staying away from work,²⁹ but their results may have been biased by low response rates.

Our data suggest a 10% incidence of ILI in respondents themselves and a 13% incidence of ILI in the household since the beginning of the pandemic. In the Behavioral Risk Factor Surveillance System survey, conducted between September 2010 and December 2010, self-reported ILI in the past month among Whites, Blacks, and Hispanics was 8%.²¹ The low rate of self-reported incidence may explain why we do not see a direct effect of Hispanic ethnicity on ILI incidence, even though we see an indirect effect via presence of children in the household and work-related inability engage in social distancing.

Among the measures of exposure we used, we found that the number of children in the household, the inability engage in social distancing, and living in an apartment building significantly predicted ILI incidence. Compared with structural factors such as household size and living in an apartment building, worksite policies are modifiable, and we focus on them in this discussion. Inability engage in social distancing—based on the inability to work at home, lack of access to sick leave and paid time off from work, and perceived job insecurity—was higher in Hispanics than in Whites. The lack of work-related disparities between Blacks and Whites likely stems from the high unemployment rate in the African American population. Because we conducted our analyses to gauge potential inequalities at the population level, we did not constrain the sample to those who work outside the home. Just as workplace policies may affect the mechanism by which populations are unequally exposed, so too may the type of employment, and this question should be examined in the future.

The small ORs we saw translated to large effects at the population level. For instance, given that 27% were at increased risk for ILI because of lack of access to sick leave, a calculation of the population-attributable risk suggests that an 8% increase in odds of ILI for each unit increase on the “inability engage in social distancing” index (95% confidence interval [CI] = 1.02, 1.14) translates to 5.0 million cases of ILI because of lack of sick leave at the job (95% CI = 1.3; 8.8 million cases) in an adult population of 232 million.

These cases account for about 13% of CDC's estimated 38 million H1N1 flu cases in adults between April 2009 and January 2010, which translates to 23 026 additional hospitalizations and 1376 additional adult deaths attributable to worksite policies.^19,30 For Hispanics, given that 40% were at increased risk for ILI because of lack of access to sick leave, an 8% increase in odds of ILI translates to 1.2 million additional cases of ILI attributable to lack of sick leave (95% CI = 0.3; 2.0 million cases) in an adult Hispanic population of 36 million.³¹ In addition, work-related inability engage in social distancing also predicted ILI in the household. This finding suggests that the burden of morbidity attributable to ILI would be even greater with additional members in the household becoming sick because of policies at their worksites.

Thus, not only do Hispanics have less access to health care once ill,¹¹ but they also are at increased risk for ILI. These facts have implications for policies: there is a need to provide better access to vaccines, drugs, and culturally competent health care providers, as well as to reduce the source of disparities in ILI incidence.

The Healthy Families Act, under consideration in the US Congress, would mandate the provision of paid sick leave for employees.^32,33 Our data lend evidence-based support to the Health Impact Assessment of the bill¹⁵ and suggest that federal mandates for sick leave would have health impacts by significantly reducing morbidity from ILI, especially in Hispanics.

This study has some limitations. First, the completion rate was 56%. However, the completion rate on our survey is normal for surveys with oversamples of minorities and respondents taking the survey in Spanish. In addition, the Knowledge Networks panel is unique in that it includes people in cell phone–only households as well as households with no phone. Our inclusion of a large proportion of minorities also allowed us to study the social determinants of ILI in racial/ethnic subpopulations. Second, we focused on self-reported ILI. Self-reports are subject to recall bias. Furthermore, we did not specify ILI symptoms in our survey, resulting in the potential for misclassification in self-reported ILI. However, we contend that the context of our survey—an outbreak of a novel virus, with unprecedented media coverage of symptoms and of the illness itself—alleviated this potential bias.

On the other hand, the strength of our data is their uniqueness, in that the survey was designed to collect information not only on self-reported ILI incidence but also on the potentially correlated social determinants of such illness. We suggest that studies be undertaken during normal flu seasons in the United States to further explore the social determinants associated with the specific symptoms of ILI and ILI incidence in general. This would also allow an examination of the generalizability of our findings to influenza seasons in general.

The indirect effect of Hispanic ethnicity on ILI incidence, through work-related inability engage in social distancing, provides a possible explanation for unequal levels of severe disease and hospitalization among racial/ethnic subgroups in the 2009 H1N1 pandemic. Therefore, resources such as sick leave—which would allow people engage in social distancing—should be equitably distributed across the US population.