Study setting
The survey was conducted from April to August 2018 in the health operational district of Moung Ruessei, Battambang province, located in northwestern Cambodia near the border of Thailand. The three surveyed (administrative) districts included 175 villages, 20 communes, and 13 health center catchment areas. The area population consisted of an estimated 202,790 individuals in 42,072 households; village sizes ranged from 139 to 3,979 inhabitants (29–822 households)[i].
Study population, survey design and sample size
This was a cross-sectional population-based survey, using a multi-stage cluster design with probability proportional to size (PPS) and random sampling of villages (using ENA software version 2011) and random sampling of households (25 per cluster). All consenting adults 18 years and above (including visitors[ii]) were eligible for inclusion in the survey. The sampling methodology enabled an oversampling of the population ≥45 years old to account for higher expected prevalence in this population.
Sample sizes were calculated using EpiInfo software, with an estimated 7% HCV prevalence among adults ≥45 years and 1.6% HCV prevalence among all adults ≥18 years, at 95% confidence, precision = 1.0% (≥45 years) and 0.8% (≥18 years), a non-response rate of 15%, and an average household size of 4.7. A total of 147 clusters were selected (123 clusters targeting the population ≥45 and 24 clusters for the population ≥18). The final sample size required 4,784 individuals (1,610 aged ≥18 and 3,174 aged ≥45), 3,628 households (577 including ≥18 and 3,051 including ≥45).
Data collection
15 teams (1 surveyor, 1nurse) administered face-to-face standardized, pre-piloted electronic questionnaires to households and individuals. Questionnaires included information on socio-demography, migration, knowledge of HCV prevention and treatment, and individual history of HCV exposure and risk factors. Data were entered and collected using electronic tablets and then exported to a secure Kobo platform.
Statistical analysis
Statistical analyses were conducted using probabilities/sampling weights calculated for each stage of the sampling: village, household and individual. The sampling stratum considered the cluster, and analysis considered the finite population correction factor.
We conducted a multivariate analysis, accounting for the sampling design, to identify risk factors for HCV serological infection among the population ≥45 years[1].
Risk factors for seropositivity identified a priori demographic variables (age, gender, occupation, education level, ID poor card status[iii]), spatial variables (health centers catchment area, distance to Moung Ruessei referral hospital, distance to the health center from the catchment area), medical variables (history of blood transfusion and blood donation, history and location/provider type for medical injections, surgery and delivery, dental and gum treatment, type of contraception, miscarriage and abortion) and behavioral variables (tattoos, piercing, IV drug use, pedicures, manicures and frequenting of barbershops). The association between the seroprevalence and the explanatory covariates was quantified by fitting a linear multivariate regression model. The multivariate analysis retained variables from the univariate analysis with p-value less than 0.2. Estimates of the regression coefficients of the model and odds ratios with their standard error are presented. In the final model, ‘unknown’ levels of medical variables (history of blood transfusion and blood donation, surgery and dental and gum treatment) were few and are recoded as ‘none’. Statistical analyses were conducted using R version 3.4.1 (R Development Core Team, 2014). Accounting for the sampling design, the survey package (Analysis of Complex Survey Samples, Thomas Lumley) version 3.34 estimated parameters, including standard errors (Horvitz-Thompson-type standard errors are used everywhere in the survey package (Lumley T, 2017, 32). Confidence interval calculations usually used the scaled Chi squared distribution for the log likelihood from a binomial distribution (Rao JNK, 1984)
The list of villages and population data was provided by the Provincial Health Department 2016 and 2017 census data. Household lists (official household registers or notebooks) were provided by chiefs of villages and updated to include any new or temporary residents.
Community engagement
Prior to the start of the survey, meetings were organized with local authorities at all levels to introduce the objectives of the survey and to discuss the timeline and request for support. y. Mobilisers (identified by the chief of each village) visited selected households prior to the data surveyto request the household’s presence.
Laboratory procedures
Sero-infection was assessed for all participants using the SD Bioline® HCV rapid diagnostic test (Labs), performed according to the manufacturer’s instructions, using capillary blood collected by fingerpick by trained nurses.
Seropositive participants were invited to the nearest health center to assess their HCV viral load; HIV and HBV diagnostics were also offered to ensure smooth linkage to care but results were not tracked. Specimens were stored and transported to the MSF laboratory in Moung Russei hospital in cold chain (2-8°C). Samples were centrifuged the same day and stored in a refrigerator (2-8°C) before their analysis within 24h. Viral load was assessed using the Xpert© HCV viral load assay with GeneXpert© Instrument Systems (Cepheid, Sweden).
Linkage to treatment and care
Patients with detectable viral load were invited to the MSF/MoH HCV program at Moung Ruessei hospital to receive their results and initiate treatment, if desired (the survey reimbursed transport costs). Besides initial and final visits at the hospital, care was provided at the closest health facility to the patient’s home.
Ethical considerations
The survey obtained household and individual written consents.This study received ethical approval from the MSF Ethical Review Board (ID: 1816), as well as the Cambodian National Ethics Committee for Health Research (NECHR; 23 February 2018 NECHR minutes).
[1] We focused on this older cohort for the risk factor analysis both because 1) in the ≥ 18 age group there were very low rates of positivity and, among those positive, insufficient numbers of affirmative responses in the respective risk factor categories to enable meaningful analysis and 2) the ≥45 years population was of primary programmatic and advocacy interest due to the assumption (justified by the results of this survey) of the elevated comparative prevalence in this population as compared to the younger cohort.
[i] Census 2016, Cambodian Ministry of Planning
[ii] Defined as any person who slept in the household the previous night
[iii] The ID poor card is a system in Cambodia to identify impoverished households eligible to receive public assistance. For the purposes of this survey, there were three possible types of ID poor card status: 1) ID poor card 1, 2) ID poor card 2 and 3) Poor letter (whereby the village chief or another local leader provided a letter confirming the impoverished status of the family). The difference between ID poor card 1 and 2 is a matter of severity of poverty levels; the poor letter is written in the absence of having an official designation as ID poor card 1 or 2, for example in sudden or unexpected circumstances