Using a hospital admission survey to estimate the burden of influenza‐associated severe acute respiratory infection in one province of Cambodia—methods used and lessons learned

Background Understanding the burden of influenza‐associated severe acute respiratory infection (SARI) is important for setting national influenza surveillance and vaccine priorities. Estimating influenza‐associated SARI rates requires hospital‐based surveillance data and a population‐based denominator, which can be challenging to determine. Objectives We present an application of the World Health Organization's recently developed manual (WHO Manual) including hospital admission survey (HAS) methods for estimating the burden of influenza‐associated SARI, with lessons learned to help others calculate similar estimates. Methods Using an existing SARI surveillance platform in Cambodia, we counted influenza‐associated SARI cases during 2015 at one sentinel surveillance site in Svay Rieng Province. We applied WHO Manual‐derived methods to count respiratory hospitalizations at all hospitals within the catchment area, where 95% of the sentinel site case‐patients resided. We used HAS methods to adjust the district‐level population denominator for the sentinel site and calculated the incidence rate of influenza‐associated SARI by dividing the number of influenza‐positive SARI infections by the adjusted population denominator and multiplying by 100 000. We extrapolated the rate to the provincial population to derive a case count for 2015. We evaluated data sources, detailed steps of implementation, and identified lessons learned. Results We estimated an adjusted influenza‐associated 2015 SARI rate of 13.5/100 000 persons for the catchment area of Svay Rieng Hospital and 77 influenza‐associated SARI cases in Svay Rieng Province after extrapolation. Conclusions Methods detailed in the WHO Manual and operationalized successfully in Cambodia can be used in other settings to estimate rates of influenza‐associated SARI.


| INTRODUCTION
Influenza is an acute viral infection and a significant contributor to global morbidity 1-3 and mortality. 4,5 While the burden of influenza has been established in some countries, it is not well understood in many others, particularly lower-middle-income countries such as Cambodia. 6 Producing national estimates of influenza-associated morbidity and mortality and understanding influenza's impact on health systems are key deliverables in the Pandemic Influenza Preparedness Partnership Contribution Implementation Plan 2013-2016, 7 established to implement a global approach to pandemic influenza preparedness and response.
The World Health Organization's (WHO) global standards for influenza surveillance are used by many countries. [8][9][10][11] Severe acute respiratory infection (SARI) surveillance intends to capture hospitalized influenza-associated severe respiratory illness cases and uses a case definition-either recommended by WHO or country-specific-to identify severe presentations of influenza-associated respiratory disease. 12 The Cambodia-based Centers for Disease Control and Prevention and Ministry of Health's (C-CDC/MOH) SARI sentinel surveillance system defines a SARI case as a hospitalized patient with measured temperature ≥38°C or self-reported fever with symptoms of cough or sore throat, and shortness of breath or difficulty breathing within 10 days of hospital admission. The WHO-recommended case definition does not require shortness of breath and/or difficulty breathing. 11 15 To better understand the impact of influenza on the Cambodia population and inform possible future vaccine policy, health officials are exploring best practices to estimate the burden of influenzaassociated SARI in the country.
While several methods have been employed by other countries, 1,2,16-20 rate calculations are not always possible because the appropriate population count is not easily ascertained. Historically, healthcare utilization surveys (HUS) have been used to determine appropriate population denominators by estimating the catchment area for a hospital through an understanding of the population's healthcareseeking behavior. [21][22][23] However, this method requires significant financial and staffing resources, involving house-to-house surveys in a random sample of representative households to inquire about healthcare facility use. 6 To assist countries in estimating influenza-associated burden who are unable to undertake an HUS, WHO developed A

Manual for Estimating Disease Burden Associated With Seasonal Influenza
(WHO Manual), 6 which outlines an alternative method-a hospital admission survey (HAS)* -for estimating population denominators using existing data sources. We present an application in one province of Cambodia using HAS methods to estimate the population of the catchment area and calculate the burden of influenza-associated SARI.
We detail our methods and provide results and lessons learned.

| Overview of the WHO Manual
The WHO Manual contains guidance for identifying and selecting data sources, reviewing available data for quality and relevance, determining the geographic area where ≥80% of the sentinel site's case-patients reside (catchment area), and using HAS methods to estimate an appropriate population denominator. It outlines how to use the HAS, catchment population, and SARI sentinel surveillance data to calculate the burden of influenza-associated SARI.

| Data sources
Existing surveillance, hospital, and population data were evaluated for suitability to estimate the rate of influenza-associated SARI.

| Sentinel surveillance at Svay Rieng Provincial Hospital (SRPH)
We used data previously collected via C-CDC/MOH's SARI sentinel surveillance at SRPH to count the number of SARI cases and influenza-

| Hospital medical records at SRPH
We used medical chart records at the sentinel surveillance site (SRPH) to assess for under-reporting of SARI cases (sensitivity analysis) to the SARI surveillance system (Figure 1: Step 2).

| Counts of respiratory hospitalizations within the catchment area
We reviewed admission and discharge registers at all public and private hospitals within the catchment area of SRPH to count the number of hospitalizations for any respiratory illness by age group for each *In Cambodia, we used the term "Hospital Admission Review" instead of "Hospital Admission

| Svay Rieng province health center population data
Svay Rieng Province obtains population counts from each health center within the province each year (Figure 1: Step 5). These counts are reported to each operational district's health center by the village leaders and are collated and reported to the provincial health department. These counts provide district-level population data by age group; however, collection practices reportedly differ by district.

| Census-collected population data
Cambodia last conducted a census in 2008 and uses birth, death, and migration rates to extrapolate population estimates by age group, gender, and province. 24 We did not use this data source as it did not contain district-level population estimates, a necessary stratification for HAS methods.

| Nationally collected health statistics
Cambodia's MOH collects data on all hospitalizations at public hospitals and national private hospitals within Cambodia. Data are stratified by age group, gender, and primary diagnosis. We did not include this data source in our calculations because validation to determine the accuracy of the data collected was not completed. Data collected through an HAS could be used to validate this data source for a future national-level burden estimation.

| Overview of burden calculation
To estimate the rate of influenza-associated SARI in 2015, we applied methods from the WHO Manual to abstract SARI sentinel surveillance data ( Figure 1: Step 1) and conducted a sensitivity analysis of the data with chart review of selected weeks ( Figure 1: Step 2). We counted respiratory hospitalizations at all hospitals within the sentinel site's catchment area, applying HAS methods from the WHO Manual ( Figure 1: Steps 3, 4, 6). We multiplied the population of the catchment area by the percent of respiratory admissions that were hospitalized at the sentinel site to obtain a population count adjusted for those who sought care at the sentinel site ( Figure 1: Step 6). We then divided the number of influenzaassociated SARI cases per age group by the adjusted population denominator ( Figure 1: Step 7) and multiplied by 100 000 to obtain 2015 incidence rates of influenza-associated hospitalizations ( Figure 1: Step 8). The steps to calculate estimates are further outlined below.
Step 1. We counted SARI cases and influenza-associated SARI cases identified by the sentinel surveillance system at SRPH in 2015, stratified by age. The WHO Manual recommends at least 1 year of SARI surveillance; we selected 2015, as that was the first year SRPH collected district-level population data on SARI cases. We followed the WHO-recommended age groups for influenza surveillance reporting for those ≥15 years, specifically: 15-49 years, 50-64 years, and ≥65 years. 11 For those <15 years of age, we used age groups that corresponded with available population data: <1 year and 1-14 years. A SARI case at SRPH was defined according to the C-CDC/MOH case definition. SRPH clinicians attempted to enroll all hospitalized patients meeting the SARI case definition from all medical, non-obstetric wards.
Nasopharyngeal swabs, oropharyngeal swabs, or sputum samples were taken from enrolled case-patients and were tested for influenza viruses at the National Institute for Public Health (NIPH) in Phnom Penh using real-time reverse transcription polymerase chain reaction (rRT-PCR).
Step 2. We conducted a sensitivity analysis at SRPH to evaluate SARI case ascertainment and to determine whether adjustments were needed to account for missing information. We compared SARI case-patients identified through the surveillance system during selected weeks (Epi Weeks 4, 24, 27, 30, 37, and 40) † to hospitalized patients identified by chart review during the same time period. From each medical record, we collected age, district of residence, admission and discharge dates, discharge diagnosis, respiratory symptoms (fever, cough, sore throat, difficulty breathing), and illness onset date.
We classified each patient as either a SARI case or not using both C-CDC/MOH and WHO SARI case definitions. We divided the number of cases identified through chart review by the number reported through the surveillance system and applied the calculated ratio as an adjustment to the numerator in our burden calculation (Figure 1: Step 3. We identified the catchment area, which is the geographic area where most case-patients seeking care at the sentinel site resided. The WHO Manual recommends that the catchment area for a sentinel site should include the administrative divisions where ≥80% of the SARI case-patients reside for the site. We used address data from the SARI surveillance system to determine the districts within the province where the majority of SARI case-patients lived.
Step 4. We evaluated respiratory diagnoses to capture the most applicable hospital admissions. While the WHO Manual recommends using a diagnosis of pneumonia to determine the number of respiratory hospital admissions, we were concerned that using one specific definition might result in missing some SARI-related admissions.
None of the hospitals within the catchment area used International Classification of Diseases (ICD) coding for admission and discharge diagnoses, so we evaluated a comprehensive list ( Table A1 in Appendix) of free-text diagnoses and compared them to recorded discharge diagnoses from both SARI case definitions to determine the most appropriate ones for future use.
Step 5. We obtained district-level population data stratified by age. To calculate the influenza-associated SARI rate, we used age-specific population data for the catchment area of our sentinel site. While we identified two sources of population data, we used the Svay Rieng Province Health † Epi weeks were defined as: Center Population data because district-level population data were necessary to match the population denominator to the catchment area.
Step 6. With our HAS, we counted respiratory hospitalizations at all hospitals within the catchment area to determine the proportion of respiratory hospitalizations at the sentinel site, relative to all respiratory hospitalizations. We manually reviewed hospital admission and discharge registers to determine the number of respiratory admissions using the predetermined list of respiratory diagnoses at each hospital within the catchment area. We recorded age, gender, district of residence, admission and discharge diagnoses, symptom onset date, and the outcome of each patient that lived within the catchment area and had at least one respiratory diagnosis associated with their hospitalization. The count of respiratory hospitalizations, stratified by age, at the sentinel site was divided by the count of total respiratory admissions at all hospitals within the catchment area for that age group. The resulting proportional number of respiratory admissions at the sentinel site was multiplied by the population of the catchment area to create an adjusted population denominator by age group for the incidence rate calculation ( Figure 1: Result B).
Step 7. We calculated the influenza-associated SARI rate using adjustment factors (Figure 1: Result C).
• Adjusted population denominator formula: population in catchment area by age group × percentage of admissions at sentinel site by age group • Incidence rate formula: (Influenza-associated SARI cases/adjusted population denominator) × 100 000 Step 8. We multiplied resulting rates by the age-specific provincial population data to obtain a provincial estimate of influenza-associated SARI cases for SRPH in 2015 (Figure 1: Result D).

| SARI cases and catchment area
During 2015, 214 SARI case-patients at SRPH were identified and enrolled in the surveillance system. Svay Rieng Province has one town and seven districts. We included Svay Rieng town and four districts, capturing 95% (203/214) of SARI case-patients in the catchment area ( Figure 2).
Of those, 8% (n = 17) of 203 specimens tested were positive for influenza (Table 1, Figure A1 in Appendix). We identified all medical facilities located within the catchment area of SRPH, which included three public district hospitals and four private admitting facilities (Figure 2).

| Population data
Using Svay Rieng Province Health Center Population Data, the Svay Rieng Province population in 2015 was 596 539 and the population within the catchment area was 487 489 (Table 2a).

| Influenza-associated SARI rate in the SRPH catchment area
We calculated the influenza-associated SARI rate using both SARI case definitions. Using the C-CDC/MOH case definition, we estimated an influenza-associated SARI rate for all ages of 6.7/100 000 persons in 2015 within the SRPH catchment population ( Table 2). After adjusting by a factor of 2.0 for possible decreased case ascertainment using the C-CDC/MOH case definition compared to the WHO case definition, we estimated an influenza-associated SARI rate of 13.5/100 000 persons (Table 2b).

| Influenza-associated SARI cases in Svay Rieng Province
Using the unadjusted influenza-associated SARI rate, we estimated 38 influenza-associated SARI cases in Svay Rieng Province in 2015 after extrapolating the rate to the provincial population (n = 596 539).
By adjusting for possible under-ascertainment of cases by using the WHO case definition and applying the resulting catchment area hospitalization rate to the provincial population, we estimated 77 influenzaassociated SARI cases in Svay Rieng Province in 2015 (Table 2b). fever with cough or shortness of breath or difficulty breathing. They estimated an incidence rate of 290-470 cases per 100 000 persons for patients aged less than five years and 20 per 100 000 persons for those aged >5 years. 25 Differences in the case definition used for enrollment could account for some of the variation in estimates between countries, in addition to factors mentioned previously.

| DISCUSSION AND LESSONS LEARNED
Our study had several limitations. We found a lower rate of influenza-associated SARI than we expected. Cambodia's case definition requires that the patient have difficulty breathing. This definition is more specific than the currently recommended WHO case definition and may have limited case enrollment; however, even after adjusting for possible low case ascertainment using C-CDC/ MOH case definition, our estimate was still low. SRPH is a relatively new sentinel site (initiated in 2014) and surveillance implementation challenges may have resulted in identification and enrollment of fewer case-patients than met the case definition. We may also have missed a large fraction of the population that either chose not to seek care for their illness (ie, community burden of disease) or sought care outside of the province. Our study could not account for these factors.
Our estimate was for one province of Cambodia, which may not be representative of Cambodia; this initial estimate was not intended to provide a national estimate. Future work to generate a national estimate is needed to understand the burden of influenza-associated SARI in Cambodia.
The HAS method described by the WHO Manual 6 has several advantages. First, it is a relatively simple method that can be used in other settings where hospitalization and population records are available. We successfully conducted data collection in 7 days with 12 data collectors and six supervisors using existing population data, existing paperbased medical records, and hospital admission and discharge registers as data sources. Second, it employs established laboratory-based surveillance systems for influenza and encourages the evaluation and strengthening of these systems to ensure the collection of high-quality data. Third, it allows health officials to better understand healthcareseeking behavior for respiratory illnesses within their jurisdiction.
Each data source used in our estimate presented challenges for use and interpretation. By applying the WHO Manual to Svay Rieng Province, we learned lessons that may be useful to other settings where the WHO Manual is being considered.
While the intent of a burden study is not to review and evaluate the surveillance system, such review may inform projects that use  hospital; therefore, an evaluation to optimize the list of included diagnoses is recommended before beginning an HAS.
Another lesson learned was the importance of identifying and evaluating all possible data sources, including population data. The HAS methodology requires reliable and accessible population data for relatively small administrative units. The availability and quality of these data must be evaluated before deciding to conduct an HAS.
We identified two sources of population data, one of which had data stratified by district, which we were able to use for our provincial estimate.

| CONCLUSION
Using HAS methods derived from the WHO Manual and several preexisting data sources, we estimated an influenza-associated SARI rate of 6.7/100 000 persons for the catchment area of Svay Rieng Hospital; after adjusting for possible decreased case ascertainment using the C-CDC/MOH case definition, we estimated a rate of 13.5/100 000 persons. We evaluated data sources, detailed steps of implementation, and identified lessons learned. A careful review of all available data sources before beginning data collection will improve the feasibility of the study and quality of the data collected. Health officials may find our operationalization of the WHO Manual methods and detailed steps helpful when working toward similar estimates in similar contexts.