Estimation of the National Disease Burden of Influenza-Associated Severe Acute Respiratory Illness in Kenya and Guatemala: A Novel Methodology

Background: Knowing the national disease burden of severe influenza in low-income countries can inform policy decisions around influenza treatment and prevention. We present a novel methodology using locally generated data for estimating this burden. Methods and Findings: This method begins with calculating the hospitalized severe acute respiratory illness (SARI) incidence for children <5 years old and persons ≥5 years old from population-based surveillance in one province. This base rate of SARI is then adjusted for each province based on the prevalence of risk factors and healthcare-seeking behavior. The percentage of SARI with influenza virus detected is determined from provincial-level sentinel surveillance and applied to the adjusted provincial rates of hospitalized SARI. Healthcare-seeking data from healthcare utilization surveys is used to estimate non-hospitalized influenza-associated SARI. Rates of hospitalized and non-hospitalized influenza-associated SARI are applied to census data to calculate the national number of cases. The method was field-tested in Kenya, and validated in Guatemala, using data from August 2009-July 2011. In Kenya (2009 population 38.6 million persons), the annual number of hospitalized influenza-associated SARI cases ranged from 17,129-27,659 for children <5 years old (2.9-4.7 per 1,000 persons) and 6,882-7,836 for persons ≥5 years old (0.21-0.24 per 1,000 persons), depending on year and base rate used. In Guatemala (2011 population 14.7 million persons), the annual number of hospitalized cases of influenza-associated pneumonia ranged from 1,065-2,259 (0.5-1.0 per 1,000 persons) among children <5 years old and 779-2,252 cases (0.1-0.2 per 1,000 persons) for persons ≥5 years old, depending on year and base rate used. In both countries, the number of non-hospitalized influenza-associated cases was several-fold higher than the hospitalized cases. Conclusions: Influenza virus was associated with a substantial amount of severe disease in Kenya and Guatemala. This method can be performed in most low and lower-middle income countries.

Siaya District Hospital (SDH) is located in Siaya District, Nyanza Province in rural western Kenya. HIV prevalence in Nyanza Province was 13.9% among adults in 2009 [11]. Population-based surveillance at SDH is embedded in a larger Health and Demographic Surveillance System (HDSS) [12]. For this analysis, a case of SARI in children <5 years was defined as a modification of the WHO definition of pneumonia as any child, residing in a defined catchment area, hospitalized with cough or difficulty breathing and any one of the following: tachypnea for age group, unable to drink or breastfeed, vomits everything, convulsions, lethargic or unconscious, nasal flaring, grunting, oxygen saturation <90%, chest indrawing, or stridor in a calm child [13],[14]. SARI for persons ≥5 years was defined as any hospitalized case with cough, difficulty breathing, or chest pain during the previous 14 days. The denominator used to calculate rates was from Karemo Division, the division located closest to SDH. No other inpatient facilities exist in Karemo. Kilifi District Hospital (KDH) is located in Kilifi district, Coast province in eastern Kenya. In 2009 among adults, Coast province had an HIV prevalence of 4.2% [11]. A HDSS is in operation within the district run by the KEMRI-Wellcome Trust Research Programme, which allows accurate determination of age-specific denominators [15],[ 16]. For most residents of this study area, KDH is the nearest inpatient facility; however, there are three other inpatient hospitals within the district. KDH only had population-based data for children <5 years for August 2009–July 2010. The same case definition of SARI in children was used as in Siaya. The base rates calculated for Kilifi were calculated using administrative sub-locations within 5 km of KDH. Sentinel hospital influenza surveillance in Kenya was established in 2007 and took place in all 8 provinces. Specimens were collected from hospitalized patients who met the SARI case definition, which for persons ≥5 years old was defined as a fever of ≥38°C with cough or shortness of breath or difficulty breathing and for children was defined as given above [13]. Influenza vaccine is rarely used in the public or private sector in Kenya, as in other sub-Saharan African countries [17]. Figure 1 provides an overview of the methodology. Data input steps are in white boxes and data output are in shaded boxes. From population-based surveillance sites with known catchment populations, we calculated the rate of hospitalized SARI, referred to as the ‘base rate.’ We used a catchment area of 5 kilometers from the district hospital as our denominator, because we felt persons within this area would be most likely to seek care at this facility and we could define the most accurate incidence, as healthcare utilization has been shown to decrease with greater distance from a health facility in Africa [18]. The number of SARI patients admitted to the surveillance hospital for each year of surveillance was divided by the age-specific hospital catchment populations. We calculated an adjustment factor for each province in the country based on known risk factors for SARI adapted from a method used to calculate the global incidence of pneumonia in children by Rudan et al., which applied adjustments for five risk factors for pneumonia in low and lower-middle income countries, including malnutrition (weight-for-age z-score <−2), low birth weight (≤2500 g), non-exclusive breastfeeding (during the first 4 months of life), household air pollution (defined as using solid fuels), and crowding (defined as ≥5 people per household) [19]. Our adjustments were made based on the prevalence of these risk factors in each province and their relative risk for childhood pneumonia as defined by Rudan et al. based on their review of the literature (Table 1) [11],[ 20]. Because of the elevated risk for hospitalized SARI in HIV-infected persons, we added an adjustment for HIV prevalence using the same equation [21],[22],[23],[24],[25],[26]. We did not have data on HIV prevalence in children in Kenya, therefore we calculated this prevalence using an algorithm which took into account the prevalence of HIV-infected mothers from antenatal clinics and the expected vertical transmission rate, accounting for penetration of prevention-of-mother-to-child-transmission programs, and transmission through breastfeeding (Appendix S2). For persons ≥5 years old, the risk factors used for adjustment were household air pollution, crowding, and HIV prevalence. Due to a lack of studies found during a literature review of these risk factors for adult pneumonia, we assumed that the relative risk of SARI for each risk factor was the same as for children. The rate of hospitalized SARI by province was calculated by applying the risk-factor adjustment described in Step 2a and further adjusting for healthcare-seeking practices using a ratio of healthcare-seeking in the base province to each other province. Since 2007, national Demographic and Health Surveys (DHS) have included a standardized question that asks about healthcare-seeking practices for acute respiratory illness (ARI) [11]. The question currently asks caretakers ‘if their children under age five had been ill in the two weeks preceding the survey with a cough accompanied by short, rapid breathing or difficulty breathing which the mother considered to be chest-related.’ If yes, the caretaker is asked if he or she ‘sought advice or treatment from a health facility or a provider’, which excludes pharmacies, shops and traditional healers. The percentage of children with ARI in the past two weeks who sought care at a health facility or provider is reported by administrative region in the country (e.g., province). As there was no question in the DHS asking about healthcare-seeking for adults and since the relative rather than absolute healthcare-seeking by province was more important for the adjustment, we used the same proportion for adults who sought care as for children from the DHS. We also assumed that healthcare-seeking for SARI in a province would be proportional to healthcare-seeking for ARI as defined in the DHS, and applied the ratio of these percentages to further adjust the rate of SARI for each province compared to the base rate province. (Appendix S1, Equation 3.) The percentage of hospitalized SARI associated with influenza A and B viruses was obtained from established sentinel influenza surveillance sites. In Kenya, each province had one sentinel surveillance site and in Guatemala only the two provinces with population-based SARI surveillance had influenza-specific data available for this analysis. In Kenya, for provinces with <25 cases of SARI during a surveillance year, we considered the estimate as unstable due to small numbers or insensitive surveillance, and so used a weighted average of the percentage of hospitalized influenza-associated SARI from provinces with ≥25 cases, weighted by the number of samples taken. We applied these percentages to the rate of hospitalized SARI by province to determine the rate of hospitalized influenza-associated SARI for each province. (Appendix S1, Equation 4.) Using data from published healthcare utilization surveys asking about healthcare-seeking practices for pneumonia, we estimated the rate of non-hospitalized SARI [27],[28],[29]. In these surveys, ‘pneumonia’ was defined as cough or difficulty breathing for more than two days or a diagnosis of ‘pneumonia’ by a healthcare worker. This estimation was based on the assumption that those patients have the same severity of illness, but did not seek care due to lack of access. We felt that compared to the DHS, the healthcare utilization survey healthcare-seeking questions were more relevant for SARI, as the questions focused on pneumonia, rather than the more nonspecific ARI question of the DHS; we expected more healthcare-seeking for more severe episodes like SARI than for all ARI. In addition, the healthcare utilization survey questions were asked for both children and adults. In both countries the healthcare utilization survey defined pneumonia as, ‘cough or difficulty breathing that lasted more than 2 days or a diagnosis of pneumonia given by a doctor or professional healthcare provider in the last year’. Respondents who reported a pneumonia episode in the past year were asked about healthcare-seeking. For this analysis, we considered the percentage of patients with pneumonia who sought care at a hospital as indicative of the percentage of persons who would have access to a hospital if they were to have an episode of influenza-associated SARI. Because healthcare utilization survey data was only available in one province in Kenya and two departments in Guatemala, we adjusted the healthcare utilization survey-derived percentage of those who sought care at a hospital for pneumonia to each province by applying the same ratio of healthcare-seeking for ARI from the DHS that we used in Step 2b; although in this case the ratio was for each province divided by the province in which the healthcare utilization survey was done (which may or may not have been the same as the province where the base rate of SARI was obtained). The age-specific provincial populations were multiplied by the rates of hospitalized and non-hospitalized influenza-associated SARI to determine the number of cases for each province. The populations of the provinces were obtained from the most recent national census data, with projected annual population growth [30],[31]. The numbers of cases in each province were summed to calculate the national number of cases. Confidence intervals were estimated by bootstrapping each data input in steps 1–4 above 1,000 times which resulted in 1,000 estimates of the number of hospitalized and non-hospitalized influenza-associated cases in the country. The upper and lower limits of the 95% confidence intervals were the 2.5th and 97.5th percentiles of these estimates, respectively. Confidence intervals are not symmetric because some inputs were re-sampled on the margins of the parameter space (e.g. proportions close to 0 or 1), reducing their potential variability in only one direction. We validated this methodology in Guatemala using the slightly different case definitions used in surveillance there. The Centers for Disease Control and Prevention (CDC), in conjunction with the University of the Valley of Guatemala (UVG) and the Guatemalan Ministry of Public Health and Social Assistance, conducted population-based surveillance for pneumonia in two departments –in Santa Rosa, a rural lowland department in southeast Guatemala, at the National Hospital of Cuilapa, and in Quetzaltenango, a semi-urban highland department in western Guatemala, at Western National Hospital. During the 2009 influenza pandemic year, immunization for pH1N1 was reasonably high, but coverage for seasonal strains is typically low [32]. At both Guatemalan sites, a case of pneumonia was defined as a hospitalized patient, residing in the hospital's pre-defined catchment area with at least one sign of an acute infection (e.g. fever, abnormal white blood cell count, hypothermia) and at least one sign or symptom of a respiratory tract illness (e.g. cough, rapid breathing, production of phlegm, chest pain, difficulty breathing) [33]. All patients meeting the pneumonia case definition were tested for influenza virus using real-time PCR [34]. Besides Quetzaltenango and Santa Rosa, no other sites had complete influenza sentinel surveillance data available. In Guatemala, there were a few deviations from the methodology described above. In Step 1, the base rate included an adjustment for the proportion of SARI cases hospitalized at non-surveillance hospitals within the catchment area. Also, there was no recent DHS data for Guatemala, but a similar survey (National Survey of Maternal and Child Health [ENSMI]) contained data on healthcare-seeking practices (Step 2b) and risk factor prevalence (Step 2a) [35]. Finally, because prevalence of HIV remains low (<1%), no adjustments were made for HIV in Step 2a. Approval of protocols and consent forms for ongoing surveillance, where relevant, were obtained by the respective ethical review committees of CDC, KEMRI and Wellcome-Trust, and UVG.

• fuller, james a.
• summers, aimee
• katz, mark a.
• lindblade, kim a.
• njuguna, henry
• arvelo, wences
• khagayi, sammy
• emukule, gideon o.
• linares-perez, nivaldo
• mccracken, john p.
• nokes, d. james
• ngama, mwanajuma
• kazungu, sidi
• mott, joshua a.
• olsen, sonja j.
• widdowson, marc alain
• feikin, daniel r.