The search yielded a total of 12,792 unique articles after duplicates were removed (Fig. 1). After reconciliation, 405 articles moved on from the first phase of screening. Moving on to the full-text review (phase II), we narrowed the diseases of interest to nine diseases: articles focusing on other diseases were still included in the review. Reviewers examined the full text of 396 articles (nine articles did not have the full text available, they were likely poster abstracts) and found that 114 articles did not present costs generated empirically and 245 articles did not present data on any of the nine diseases of interest. We made a final selection of 37 articles. The level of inter-reviewer agreement (Cohen’s kappa) was 68.1% (substantial agreement) by the end of the screening process .
Countries, Illness, and Populations
From the 37 articles, we drew 267 different sets of costs. Five of the nine VPDs were identified from the 267 sets of costs: GE (116 sets, 43%), pneumonia (121, 45%), meningitis (22 sets, 8%), JE (three sets, 1%), influenza (two sets, 1%), and other illnesses (three sets, 1%). In the latter, two sets only specified “other” and “very severe disease” related to pneumococcal infection and one specified acute otitis media. We found no estimate for hepatitis B, measles, rubella, and YF in children. Children between the age of 1 and 59 months were the main age group under study (218 sets, 81%), followed by children aged 0–3 years (35 sets, 13%). Children aged older than 5 years were also included in six sets of costs, 2% (Table 1).
The largest portion of these costs came from countries in Sub-Saharan Africa (97 sets, 36%), followed by South Asia (80 sets, 30%), East Asia and Pacific (67, 25%), Latin America and the Caribbean (16 sets, 6%), Europe and Central Asia (four sets, 2%), and the Middle East and North Africa (three sets, 1%). Based on Gavi’s 2018 Annual Progress Report , most sets came from countries in preparatory (75 sets, 28%) and accelerated transitions (also 75 sets, 28%), followed by countries initiating self-financing (59 sets, 22%) and non-Gavi-eligible countries (51 sets, 19%). Indonesia was the only fully self-financing country with a COI study (seven sets, 3%).
Most COI evidence was generated in countries where publicly funded healthcare was available. The most commonly reported perspectives were the government perspective where we found 96 sets of costs (36%) and the household perspective (79 sets, 30%) for which most of the data collection was performed in public healthcare facilities. Studies that adopted the healthcare and the household perspectives also adopted a societal perspective in 47 sets of costs (18%). Thirty sets of costs (11%) were reported from the perspective of private healthcare providers without further details on whether the costs were transferred to users or other sources of revenue. Eleven sets of costs (4%) took a healthcare provider perspective, not differentiating what was paid by the government and by the private sector. One article reported costs borne by health insurance, hence taking a third-party payer perspective in four sets of the costs (2%).
Most sets of costs were associated with COI in urban settings (127 sets, 48%). Eighty-three sets (31%) were associated with rural settings and 57 sets with mixed urban and rural settings (21%). In our selected articles, each set of costs could combine costs for more than one facility level: 164 sets of costs (61%) included costs from tertiary healthcare facilities, 81 (30%) from secondary healthcare facilities, and 93 (35%) from primary healthcare facilities.
Scope and Methods
Of the 37 articles, 11 (30%) integrated a COI component with primary data collection within a larger study. Five of 37 papers were cost-effectiveness analyses, four were burden of disease studies, and two were randomized controlled trials. The remaining 26 articles were stand-alone COI studies. These COI studies took an incidence-based cross-sectional approach, defining the COI around the healthcare facility visit for an acute episode of the illness. While they focused on the acute episode of the illness, most sets of costs adopting the household perspective included a follow-up period, 7–14 days after the initial episode.
To identify VPD cases, 23 articles reported costs with laboratory-confirmed cases (GE, meningitis, JE) and nine with radiology-confirmed cases (pneumonia). Among these, six studies used both laboratory testing and radiology. Eleven articles relied on clinical assessment alone to identify cases.
Twenty-one articles (57%) took a prospective approach to data collection, seven (19%) took a retrospective approach, and nine (24%) combined prospective and retrospective approaches. Costs from the household perspective were always estimated prospectively as these relied on caregiver responses. Ten articles (27%) included caregiver interviews. Most of them (eight articles, 22%) performed a follow-up interview 7–14 days after the initial interview. One article performed interviews 6 weeks later , another 5–12 months afterward .
Among the articles assessing the cost of GE (19 articles), half of them used an established costing method. Eight used the WHO guidelines to estimate the economic burden of diarrhea published in 2005 [14, 16,17,18,19,20,21,22], one used the WHO guideline for cost analysis in primary healthcare published in 1994 , and one used unpublished WHO guidelines cited as “WHO (U Griffiths, R Rheingans, D Walker, unpublished data)” . One article focusing on pneumonia and meningitis used the national guideline from the Ministry of Finance (Vietnam) to estimate capital costs . None reported a qualitative component (expert consultations) to identify potential costs for households and the healthcare system.
Types of Costs
All 37 articles presented direct medical costs. Twenty-three (62%) also presented direct non-medical costs and 24 (65%) presented indirect costs. Twenty-two articles presented all three types of costs. Direct medical costs included medications and medical procedures, and most articles presented such medical costs aggregated.
Five articles (14%) included overhead costs [26,27,28,29] and two (5%) added discounted capital costs [25, 29]. Four articles [25,26,27, 29] counted them as part of the medical costs, while one  included them as part of an “indirect cost” from the provider perspective.
Direct non-medical costs included the cost of transportation (21 of 37 articles), meals during the facility visit (9 of 37), caregiver accommodations (3 of 37), and other costs related to childcare (8 of 37) such as diapers, visitors’ gifts, and transportation for non-caregivers. Four articles presented aggregated direct non-medical costs only and one article aggregated all direct medical and non-medical costs as a “total cost of admission”.
Indirect costs were based on income loss for the caregiver(s) (18 of 37 articles) and time loss for the caregiver because of disability (1 of 37 articles). The other four articles presented only an aggregated indirect cost value and did not describe cost composition. All indirect costs were estimated through a human capital approach, considering the productivity loss of caregivers. Among the articles reporting income loss, six articles estimated income loss for the surveyed caregiver only [15, 17, 24, 26, 30, 31], eight estimated it for both the father and the mother of the sick child [18, 19, 22, 32,33,34,35,36], and three assessed income loss for all reported caregivers [21, 27, 37]. One article did not disclose whose income was lost .
Most articles reported the costs during the year of data collection or the year following data collection, without any correction for inflation. One article  used unpublished costs originally collected in 1998–2000 and adjusted them to the 2010 national currency value. Several articles [18, 19, 24, 38,39,40] collected costs over several years and did not specify the year of the currency value; we assumed the year of the currency was the starting year for data collection to correct for the missing base year.
Direct medical, non-medical, and indirect costs per episode for inpatient care were greater than for outpatient care for the household perspective (Table 2). There was no apparent trend between these costs and country income status. For pneumonia and GE episodes, household direct medical costs had a range of $3.52–$125.39 per hospitalized case and $0–$53.87 per ambulatory case. For JE, the medical costs ranged between $577.65 and $1268.84. Non-medical costs were similar across country income statuses and diseases, $1.21–$28.29 (inpatient), and $0–$8.94 (outpatient), with an exception for meningitis where they were much higher: $28.78–179.46 (mixed inpatient/outpatient). Ranges of reported indirect costs were similar between diseases and overlapped across country income statuses with larger variations with higher income statuses: $11.21–$41.31 (low income), $2.25–$90.78 (lower-middle income), and $0.55–$214.55 (upper-middle income) (Tables 2, 3, and 4).
In addition to differences by the type of care provided, government costs per hospitalized episode increased with higher country income status (Table 3). Governments spent $46.76–$84.95, $130.86–$442.54, and $205.07–$6623.99 per hospitalized episode across all diseases in low-, lower-middle-, and upper-middle-income countries, respectively. In most articles including both the household and the government perspectives, governments faced greater costs per episode of illness than households. In Le et al., the government spent more than households on medical care for an episode of pneumonia and meningitis in Vietnam; however, when including non-medical costs, the household direct costs exceeded those of the government . There were strong differences in societal costs across diseases and types of care (Table 4).
We examined the share of the costs from the household perspective and focused on 27 sets of costs (from nine articles) that looked at households using public healthcare facilities, who reported direct medical, direct non-medical, and indirect costs (Appendix 2 in ESM). Across diseases and settings, the proportion of non-medical and indirect costs outweighs the medical costs. Non-medical costs dominate (54%) other costs over outpatient care, while indirect costs take the highest share (43%) of the total cost for inpatient care. Medical costs made 21% of the cost of hospitalized pneumonia and between 1% and 14% for outpatient pneumonia, while non-medical and indirect costs estimated at 62–86% and 10–15%, respectively. For GE, medical costs made 12–44% of the cost of a hospitalized case with non-medical (14–16%) and indirect costs (41–72%). One set reported that households had no out-of-pocket expenses for outpatient GE, facing only indirect costs.
Three sets of costs included all three types of costs for households using private healthcare facilities, representing inpatient care only. The average proportion of direct medical costs (69%) is higher than that of direct non-medical (16%) and indirect costs (15%) (Appendix 2 in ESM).
Economic Burden on Households
In addition to the breakdown of the burden of costs borne by the households, we examined whether authors took the additional step to interpret COI estimates in relation to income or total expenditure of the household or government, which we defined as economic burden measures.
In this review, 18 articles (60%) described the economic burden of illness with 45 economic burden measures. These composite measures contain a variety of numerators and denominators as demonstrated in Table 5.
The result shows that the percentage of COI as a percentage of household expenditure ranges from 43% to 83%. Cost of illness as a percentage of household income ranges from 0.39% to 1000%, depending on the illness and the choice of numerator and denominator. Cost of illness as a percentage of government per capita expenditure falls between 3.5 and 82%.
The main source of funding for the studies identified in this review came from multi-lateral agencies (WHO, UNICEF), non-government organizations, and philanthropies (22 articles, 59%), followed by governments and public organizations (16 articles, 43%) and the private sector (eight articles, 22%). Four articles did not disclose any source of funding [27, 28, 34, 35].