Impact of H1N1 on Socially Disadvantaged Populations: Systematic Review

Background The burden of H1N1 among socially disadvantaged populations is unclear. We aimed to synthesize hospitalization, severe illness, and mortality data associated with pandemic A/H1N1/2009 among socially disadvantaged populations. Methods/Principal Findings Studies were identified through searching MEDLINE, EMBASE, scanning reference lists, and contacting experts. Studies reporting hospitalization, severe illness, and mortality attributable to laboratory-confirmed 2009 H1N1 pandemic among socially disadvantaged populations (e.g., ethnic minorities, low-income or lower-middle-income economy countries [LIC/LMIC]) were included. Two independent reviewers conducted screening, data abstraction, and quality appraisal (Newcastle Ottawa Scale). Random effects meta-analysis was conducted using SAS and Review Manager. Conclusions/Significance Sixty-two studies including 44,777 patients were included after screening 787 citations and 164 full-text articles. The prevalence of hospitalization for H1N1 ranged from 17–87% in high-income economy countries (HIC) and 11–45% in LIC/LMIC. Of those hospitalized, the prevalence of intensive care unit (ICU) admission and mortality was 6–76% and 1–25% in HIC; and 30% and 8–15%, in LIC/LMIC, respectively. There were significantly more hospitalizations among ethnic minorities versus non-ethnic minorities in two studies conducted in North America (1,313 patients, OR 2.26 [95% CI: 1.53–3.32]). There were no differences in ICU admissions (n = 8 studies, 15,352 patients, OR 0.84 [0.69–1.02]) or deaths (n = 6 studies, 14,757 patients, OR 0.85 [95% CI: 0.73–1.01]) among hospitalized patients in HIC. Sub-group analysis indicated that the meta-analysis results were not likely affected by confounding. Overall, the prevalence of hospitalization, severe illness, and mortality due to H1N1 was high for ethnic minorities in HIC and individuals from LIC/LMIC. However, our results suggest that there were little differences in the proportion of hospitalization, severe illness, and mortality between ethnic minorities and non-ethnic minorities living in HIC.


Introduction
In 2009, a novel H1N1 influenza virus strain circulated, which gave rise to the 2009 H1N1 pandemic (influenza A/Mexico/2009 (H1N1)). The H1N1 pandemic was associated with a high burden of illness in terms of hospitalizations, severe illness, absenteeism, and cost. For example, in the United States (US), over 43,677 laboratory-confirmed cases of pandemic H1N1 2009 were reported [1]. Using a mathematical model, it was estimated that between 1.8 million to 5.7 million cases occurred, including 9,000 to 21,000 hospitalizations in the US [1]. In Spain, the average work absenteeism due to the 2009 H1N1 pandemic ranged from 9 to 30.5 days [2]. The economic burden for those with confirmed influenza was estimated as J144,773,577 in Spain (95% confidence interval, CI: 13,753,043-383,467,535). In Italy, the estimated economic burden due to laboratory-confirmed H1N1 2009 pandemic ranged from J1.3 to J2.3 billion [3]. In Australia, the economic burden of treating H1N1-admitted patients to the intensive care unit (ICU) was over AU $65,000,000 [4]. These estimates indicate that a significant burden of illness was observed due to the 2009 H1N1 pandemic.
Previous reviews of the 2009 H1N1 pandemic reported that the majority of cases occurred among young to middle-aged adults often in those without comorbidity, followed by children and adolescents [5,6]. Individuals with a greater burden of illness included the elderly, obese individuals, pregnant women, or those with comorbidity [5,6]. In addition, it has been hypothesized that greater burden of illness was associated with poverty and individuals without access or disproportionate access to healthcare [7]. To examine this further, we aimed to synthesize hospitalization, severe illness, and mortality data associated with pandemic A/H1N1/2009 among socially disadvantaged populations, in-cluding low socioeconomic status, ethnic minorities, groups without access or disproportionate access to healthcare, and lowincome economy countries or lower-middle-income economy countries [LIC/LMIC].

Methods
A systematic review protocol was used to guide the methods of our review, based on the Preferred Reporting Items for Systematic Reviews and Meta-analysis (PRISMA) Statement [8]. Our research question was: ''what is the evidence that the burden of H1N1 was associated with social disadvantage?'' At the time of study conduct, a similar systematic review protocol focusing on H1N1 burden among the socially disadvantaged did not exist.

Search
An experienced librarian (Perrier) developed the search strategy using medical subject headings (MeSH) and text words. The MEDLINE (OVID interface, 2009 to July 25, 2011) and EMBASE (OVID interface, 2009 to July 25, 2011) electronic databases were searched to identify potentially relevant material. The full search strategy for MEDLINE is presented in Appendix S1, which was modified for EMBASE (available upon request). Targeted searching for low-income and lower-middle-income economies was also conducted in PubMed using the terms ''H1N1'' and ''country of interest''. The Eurosurveillance Journal and the Centers for Disease Control and Prevention (CDC) Morbidity and Mortality Weekly Report were hand-searched from January 2009 until August 1, 2011. The reference lists of included studies or relevant reviews [5,6,9] were scanned and a list of included studies was circulated to members of the World Health Organization (WHO) Influenza Programme to ensure all potentially relevant studies were identified.

Study Selection and Characteristics
Studies reporting the burden (prevalence of hospitalization, severe illness, and mortality) of influenza A/Mexico/2009 (H1N1) among socially disadvantaged populations (e.g., ethnic minorities, low socioeconomic status, groups without access or disproportionate access to healthcare, LIC/LMIC) were included. It was determined a priori in discussion with the World Health Organization (organization that commissioned this systematic review) that burden would include hospitalization, severe illness, and mortality. H1N1 had to be laboratory-confirmed through polymerase chain reaction, viral culture, or antibody assay [10], as each laboratory test has various advantages and limitations [11]. For example, antibody assay can detect infections missed by other laboratory methods but the diagnosis of influenza is retrospective and it takes weeks to retrieve the results [11]. According to the Center for Disease Control and Prevention (CDC), the preferred method of pandemic H1N1 2009 influenza ascertainment was polymerase chain reaction and viral culture [12]. Rapid influenza diagnostic tests were not recommended by the CDC (and hence, were excluded from this systematic review), as their sensitivity is low [10].
Ethnic minorities were classified as non-predominant races (e.g., non-Caucasians in predominant Caucasian continents, such as Europe and North America), as well as indigenous populations (i.e., first settlers in a particular territory [9]). LIC (gross national income #$1,005) and LMIC (gross national income $1,006-$3,975) were categorized according to the World Bank's classification of countries [13]. We planned to classify low socioeconomic status as high-school education or less or below the particular country's poverty line, but we did not identify articles relevant to this type of social disadvantage [14]. Inclusion was not limited by study design, publication status or language. Authors of conference proceedings were contacted to obtain the conference presentation or unpublished work. Two reviewers independently screened the titles and abstracts from the literature search and potentially relevant full-text articles for inclusion using the standardized eligibility criteria. Conflicts were resolved by discussion amongst the team.

Data Abstraction
A draft data abstraction form was developed, pilot-tested, and modified as necessary. Two reviewers abstracted all of the data independently. Conflicts were resolved by discussion amongst the review team. The following data were abstracted: study characteristics (e.g., study design, country of conduct, time period), patient characteristics (e.g., mean age, percent gender, type of social disadvantage examined) and outcomes (number of hospitalizations, severe illness, deaths). Authors were contacted for further information when the data were not clearly reported. In some instances, multiple studies reported H1N1 data from the same source (i.e., companion reports). When this occurred, the report with the most outcomes of interest or largest sample size was included and the other(s) was used for supplementary material only.

Validity Assessment
All relevant studies were assessed for risk of bias using the Newcastle Ottawa Scale (NOS) [15]. The NOS evaluates nonrandomized studies such as case-control and cohort studies and consists of 3 domains: selection, comparability, and exposure. A full explanation of the NOS can be found in Appendix S2.

Quantitative Data Synthesis
Random effects meta-analysis [16] was conducted to determine the prevalence of hospitalization, ICU admission, and mortality that occurred in high-income economy countries (HIC) and LIC/ LMIC, separately, as well as for the proportion of ethnic minorities and non-ethnic minorities experiencing these outcomes in HIC. Confounding was examined through meta-analyses of the proportion of patients with comorbidity, pregnancy, and obesity for ethnic minorities versus non-ethnic minorities in HIC. Statistical heterogeneity was examined using the I 2 and x 2 statistics [17]. Analyses were conducted in Review Manager Version 5 [18] and SAS (SAS 9.1 software, SAS Institute Inc., Cary, NC, USA).

Patient Characteristics
Some studies examined differences in the number of hospitalizations, severe illness, and deaths between ethnic minorities and non-ethnic minorities, including Caucasian, European descent, and non-Indigenous populations, which were used as the reference group ( Table 2). The number of H1N1-infected individuals ranged from 136 to 5,106 in the included studies (Tables 2 and 3). Few studies reported the percentage of female participants; in those that did report this, it ranged from 24% to 100% (Tables 2 and 3).

Methodological Quality
The majority of the studies used a sample truly representative of the average individual infected with H1N1 in the community (e.g., population-based sample of an entire province or state) or somewhat representative sample of the average individual infected with H1N1 in the community (e.g., pregnant woman from an entire state or province). Ten studies did not use a representative sample; one included volunteers [45], one included only those with severe illness [33], four included individuals with severe illness who died [20,29,37,58], two included pediatric cases in India [69,79], and two obtained data from one local hospital [43,49] (Table 4). All of the studies selected the non-exposed cohort from the same community as the exposed cohort and used a structured interview (i.e., self-report) to assess exposure (e.g., social disadvantage), except for studies conducted in LIC/LMIC, for which the data were obtained from hospital records. All of the studies ensured that the patients were not severely ill, hospitalized or dead at the start of the study, except for four studies [20,29,37,58]. All studies assessed the outcomes using record linkage, and the duration of follow-up was deemed appropriate in all of the studies. Most studies did not control for important factors, such as comorbidity (13 studies controlled for comorbidities [22,29,32,33,41,43,48,49,56,58,60,61,63]) and 18 studies had greater than 10% of the patients lost to follow-up or did not describe the follow-up rate [21,22,29,45,48,56,58,60,62,63,65,[67][68][69]72,73,79,80].
All of the studies conducted in LIC/LMIC reported hospitalization data for individuals with H1N1 ( Table 3). The prevalence of hospitalization among those infected with H1N1 was 32% in Guatemala [64], 11% in Morocco [66], and 45% in India [72]. Prevalence meta-analysis was not possible because only one study was conducted in each country.
One study of H1N1-infected pregnant women observed a significantly greater proportion of ICU admissions among nonethnic minority hospitalizations (35%) versus ethnic minority hospitalizations (35%; OR based on this study: 0.47, 95% CI: 0.28-0.79) [28]. Eight studies reported ethnicity data on the proportion of patients admitted to the ICU among those hospitalized in HIC and were meta-analyzed [25,26,30,35,44,51,59,62]. The prevalence ranged from 6-68% among ethnic minorities and 7-76% among non-ethnic minorities. One study was not included in the meta-analysis due to concerns about confounding and inclusion of a non-representative sample [43]. Excluding this study, there were no differences in ICU admissions among hospitalized non-ethnic minorities compared to ethnic minorities (n = 15,352 patients, OR 0.84, 95% CI: 0.69, 1.02, I 2 = 51%, Figure 3).

Discussion
We conducted a systematic review on the occurrence of hospitalization, severe illness, and mortality among socially disadvantaged groups of the population. The prevalence of hospitalization among those with H1N1 ranged from 24-27% among ethnic minorities and 17-87% among non-ethnic minorities in HIC and was 11-45% in LIC/LMIC. The proportion of patients admitted to the ICU among those with H1N1 ranged from 2-12% among ethnic minorities and 5-8% among nonethnic minorities in HIC and was 0-9% in LIC/LMIC. The proportion of ICU admissions among hospitalized patients ranged from 6-68% among ethnic minorities and 7-76% among nonethnic minorities in HIC. The pooled prevalence of ICU admission among hospitalizations in India was approximately 30%. The prevalence of mortality ranged from 1-21% among ethnic minorities and 4-25% among non-ethnic minorities in HIC and the pooled prevalence of deaths among hospitalizations ranged from 8-15% in India. These estimates suggest that the burden of H1N1 was significant across LIC/LMIC and HIC.
Our results are similar to previous reviews on the global burden of H1N1. In a systematic review of H1N1 in the Southern Hemisphere, 17-45% of laboratory-confirmed H1N1 cases were hospitalized, of which 8-26% were admitted to ICUs and 14-22% died [5]. A similar burden was observed in the Northern  Hemisphere, with 94% of patients hospitalized, 36% of patients admitted to ICU, and 39% died [6]. The current review is more comprehensive, with a scope encompassing all regions of the world and comparisons drawn across ethnic minorities in HIC and individuals from LIC/LMIC.
Another study pooled data on risk factors for acquiring H1N1 pandemic and included data from governmental surveillance programs across 19 countries [81]. In this study, ethnic minorities had a higher risk of hospitalization and fatality compared to the general population in Canada, Australia, and New Zealand. For severe H1N1 pandemic, ethnic minorities were under-represented among cases in Thailand and Mexico [81]. Meta-analysis results were not reported due to the small number of countries reporting data.
Some of the included studies compared the burden of H1N1 for ethnic minorities versus non-ethnic minorities in HIC, allowing comparisons between these groups. In two large North American studies, there were significantly more hospitalizations among ethnic minorities versus non-ethnic minorities [25,38]. However, there were no differences in ICU admissions or deaths among patients hospitalized with H1N1 for ethnic minorities and nonethnic minorities among studies conducted in North America and Australia. It is possible that non-ethnic minorities had a greater proportion of comorbidities, pregnancy or obesity -known risk factors for pandemic H1N1 (i.e., confounding variables) [5,6]. However, we did not find any differences in these factors between ethnic minorities and non-ethnic minorities in HIC across some of the studies that were included in the meta-analyses.
Pregnancy is a recognized risk factor for seasonal influenza and pandemic influenza [5,6,82]. One of the studies included pregnant women infected with H1N1 and the results were inconsistent across the outcomes examined [28]. For example, significantly higher hospitalizations were observed for pregnant ethnic minorities versus pregnant non-ethnic minorities, yet a significantly higher proportion of deaths occurred among pregnant non-ethnic minorities versus pregnant ethnic minorities [28]. These results could be because non-ethnic minorities had a greater proportion of comorbidity, but this information was not reported. Only one study reported this data so we were unable to examine pregnancy further through meta-analysis.
A small number of seniors were infected with the pandemic A/ H1N1/2009 virus, which is inconsistent with previous influenza pandemics [5]. This is likely because seniors were previously exposed to a similar H1N1 strain [83], affording some protection against the 2009 pandemic. However, the elderly still experienced high hospitalization and death rates [5,6]. We were unable to examine this as none of the included studies reported data specific to elderly socially disadvantaged individuals.
There are limitations associated with the conduct of our systematic review. First, we were unable to include some studies if they did not provide a breakdown for the outcomes of interest by ethnicity. Furthermore, we found that because the classification of ethnicity varies by region, drawing comparisons across countries was difficult [84]. Second, some potentially relevant studies were excluded if they did not isolate confirmed cases from suspected cases in their analysis. Third, some of the included studies did not report details, such as total number of individuals with H1N1, record of potential confounding variables or number lost to followup. For these reasons, the systematic review conduct was challenging.
It is possible that relevant unpublished studies were omitted, although efforts were made to contact authors and request access to data that had been presented at conferences. Through these efforts, we successfully obtained the results from two unpublished studies [33,35]. As well, it's worth noting that due to the small number of studies included in the meta-analysis of HIC, we were unable to assess the impact of publication bias on our results (i.e., through the use of a funnel plot). Finally, our results are generalizable only to ethnic-minorities in HIC and individuals in LIC/LMIC, as none of the identified studies reported data for other types of social disadvantage (e.g., groups without access or disproportionate access to healthcare or individuals of low socioeconomic status).
To conclude, the prevalence of hospitalization, severe illness, and mortality due to H1N1 was high for ethnic minorities in HIC and individuals from LIC/LMIC. In addition, there was an increased proportion of hospitalization among ethnic minorities compared to non-ethnic minorities in two studies conducted in North America. However, a similar risk of ICU admission and death was observed among ethnic minorities and non-ethnic minorities in studies conducted in Canada, the United States, and Australia. These results suggest that there was little difference in H1N1 burden between ethnic minorities and non-ethnic minorities living in HIC.