The economic burden of dengue fever in the Kingdom of Saudi Arabia

Rapid urbanization, global trade, and the exceptionally great numbers of worldwide visitors during Hajj and Umrah have all placed the Kingdom of Saudi Arabia at a significant risk of introducing several vector-borne tropical diseases, such as dengue fever virus (DENV) infection. In this study we estimated DENV infection cost of illness (COI) in Saudi Arabia in the period 2013–2017, by processing national data including all declared cases recorded in referral centers in the western region, being the endemic region of the country. Using a statistically validated predictive model that was built on a representative sample of 717 laboratory-confirmed cases of DENV infection, direct costs, due to care-related expenditures, were estimated by applying the predictive equation to national data. However, indirect costs, which are due to productivity loss, were estimated using the human capital model based on gross domestic product adjusted for invalidity duration. Further, under-reporting was adjusted by using an expansion factor EF = 3. We observed highest estimated costs in 2016 with over US$168.5 Million total costs, including direct (US$29.0 Million) and indirect (US$139.5 Million) costs, for a total 4415 confirmed cases. The total DENV COI for the five years was estimated as US$551.0 Million for a total 15,369 patients (59.7%) out of 25,745 declared cases, resulting in an average cost of US$11 947.6 by patient. Depending on the year, productivity years loss costs accounted for 63.3% to 83.8% of the estimated total costs. Dengue has a substantial local economic burden that costs US$110.2 Million per year, stressing the urgent need for an effective national prevention strategy to perform considerable cost-savings besides reducing morbidity.


Worldwide epidemiology of dengue
Vector-borne diseases constinue to be a major global problem, causing a significant health economic burden notably in the Eastern Mediterranean Refgions (EMR) of the World Health Organization (WHO) [1]. The global incidence of dengue virus [DENV] infection has evidenced a dramatic increase in the recent two decades and approximately 9,000 deaths are reported annually [2][3][4][5]. Therefore, the biological, epidemiological and preventive aspects of the disease have gained the attention of policy makers and funding agencies; while, the economic implication of the disease has been only recently acknowledged.

Epidemiological and historical review of dengue fever in Saudi Arabia
In Saudi Arabia, the first case of dengue fever (DF) was reported in October 1993, in Jeddah city; and since then, a national surveillance program was implemented to monitor the disease incidence [6]. In 1995, several cities in the Western and Southern regions with marked agricultural activities and rainy conditions evidenced a DF outbreak [7]. In 2001, DF was considered to be endemic in these regions comprising the cities of Jeddah, Makkah, Madinah, and Jizan [8][9][10][11], consistent with the geographical distribution of Aedes aegypti, the locally identified vector of the disease [12][13].
While recent studies (2013-2017) based in Saudi Arabia provide insufficient data about DF in several regions, Jeddah is revealed to be the major contributor to increasing the total yearly number of dengue cases in the country [8,[14][15][16][17]. Modeling techniques revealed that 15% of Jeddah districts were at high and 22% were at medium risk of DF [18]. Increased humidity, high temperature, and increased numbers of visitors during the annual pilgrimage (Hajj) and minor pilgrimage (Umrah) are all potential favoring conditions for disease transmission in Jeddah and the neighboring regions [9].

Economic burden of dengue
A small number of studies estimated the global economic burden of dengue, having multiple limitations and conflicting findings; while large methodological variations are observed in regional and multinational estimates. These studies considered various expenses and used different estimation methods, data sources and economic models, which impedes the comparison of their outcomes [5,[19][20][21]. It is expected that the quality and availability of surveillance data in several countries are based on the distribution of treatment settings and the included estimates might be based on extrapolated data. On the other hand, the economic cost of DF has no distinct profile given the variable presentation of the disease. Generally, the global economic cost of DF is likely to be higher than other major infectious diseases, including cholera, rotavirus gastroenteritis, Chagas, and canine rabies; and the main outline indicates that the burden is remarkably enormous in low-income and middle-income countries [4].

Methods used in economic studies
Many methodological techniques can be used to investigate disease costs, including cost consequence analysis (CCA), cost minimization analysis (CMA) and cost-of-illness (COI) analysis. COI studies evaluate the direct and/or indirect costs of a disease or its risk factors. These disease-oriented approaches provide plausible and efficacious analysis in the case of dengue rather than other intervention-relevant approaches [22]. It is worthy to note that the studies that employ a COI approach may be based on prevalence-based costs (dengue-attributable costs in all individuals or a representative sample in a given year) or incidence-based costs (the values of lifetime costs of newly diagnosed patients with dengue).
As with other infectious diseases, the diagnostic and management approaches represent the main costly processes in dengue. According to recent guidelines [23], dengue has no specific medication or antiviral drug for management since the targeted drugs that act directly against DENV are still in the pipeline [24]. Therefore, treatment of symptoms remains the standard approach and hospital admission may be required, notably in intensive care, for severe cases, which may entail high costs in addition to other management aspects such as laboratory diagnosis tests and further investigations that may be necessary to rule out differential diagnoses [25][26][27][28][29]. Additionally, indirect costs related to loss of productivity can emerge from decreased working days of patients or their families (caregiver or companion), travel, absence from schools or activities, etc.; which may account for up to 60% to 70% of the total dengue costs, especially in cases of mortality [30][31][32]. However, such indirect costs may be partly underestimated or even difficult to estimate because they are often incurred by the patients or their families while other direct costs are readily available and could be provided by the relevant agencies or institutions.

Rationale
In light of the remarkable increase in disease burden in Jeddah and the western region of Saudi Arabia and given the need to conduct more reliable studies for effective future interventions, it is becoming crucial to investigate the DF economic burden to measure its impact on both the patients and society. Quantifying such economic burden is indispensable for policymakers to allocate relevant resources, tailor specific prevention and control strategies, set priorities, and assess the cost-effectiveness of particular interventions [33-34].

Aim & objectives
The present study aims at providing economic data that helps political efficiency and resource prioritization for dengue prevention programs, by assessing the economic burden of the disease over the period 2013-2017.
The following objectives have been achieved: • To calculate the actual direct and indirect costs of dengue fever in a representative sample of patients; • To analyze factors and predictors of the costs and build up a predictive model for dengue direct costs based on significant and relevant predictors; • To estimate direct costs on national data 2013-2017 by applying the predictive models; • To calculate indirect costs of dengue by estimating productivity loss costs using econometric models based on the gross domestic product (GDP). The initial design consisted of a two-phase method for data collection and analysis, namely: Phase I: calculation of actual direct and indirect costs on the representative sample of patients (Source Database); Phase II: estimation of direct and indirect costs on national data for the years 2013-2017 (5 datasets), which was provided by the Public Health Department, at Jeddah, and is available at the Dengue Fever Operation Room Database, covering all governmental and private hospitals in Jeddah district (Target Database).

Ethic statement
However, estimation of indirect costs could not be done following this method due to substantial absence of relevant data in both Source and Target Databases. Thus, only direct costs were estimated using the two-phase Bottom-Up method; while the estimation of indirect costs used an alternative method, namely the Human Capital Model. The two methods are illustrated summarily in a flowchart (Fig 1) and described separately below.
reviewing the medical records in the participating hospitals; 2) the number of units or episodes for each intervention (i.e., number of hospitalization days, clinic visits, units administered for each medication, etc.) for each patient over the management period; 3) tariffs applied for each intervention, with respect to the prices applied in each center: these data were provided by the economic officers from each hospital and completed by reviewing the Saudi Food and Drug Authority (SFDA) list of medications costs [35]; 4) sociodemographic and clinical factors that   [MoH] or governmental, other governmental, averagestandard private, and high-standard private); in addition to hospitalization mode (ambulatory versus hospitalized) and mortality. Furthermore, participants were contacted for phone interviews to collect data on eventual out-of-pocket health expenses incurred during the same disease period, as well as any eventual past care costs including consultations, hospitalizations, medications and ER visits prior presentation to the participating center.
Direct costs of the sample (N = 717), which represent the outcome of Phase I, were calculated for each patient by summing up the unitary costs of each intervention multiplied by the number of episodes (e.g. hospitalization cost = unit cost of 1-day hospitalization � length of stay [LOS]). Specific assumptions were considered for the calculation of emergency room consultations costs (visit duration <24 hours was computed as 1 visit; >24-48 hours as 2 visits, >48-72 as 3 visits, etc.); fluid costs (if volume not specified, then 1-liter container was assumed by administration); antipyretics (if not specified, assumed to be prescribed as TID for 3 days); discharge medication (assumed to be 2-day prescription of paracetamol and omeprazole). According to the type of intervention, direct costs were divided into 4 sub-categories: a) Admissions costs (costs related to emergency room visits, hospital ward or intensive care unit stay, outpatients, etc.); b) Investigations costs (imaging and laboratory workups); c) Care costs (physician visits, nursing, fluids, medications, extra diet, etc.; and d) Past Direct Costs, corresponding to any care costs incurred prior to admission and diagnosis. Total direct costs were calculated as the sum of all intervention costs and was analyzed as the dependent variable in Phase II. Data management, processing and calculations in Phase I were done using Microsoft Excel 2013 for Windows.
Phase II. This Phase consisted in building the predictive model for direct costs of dengue, from the representative sample (N = 717), as a function of the relevant demographic and clinical explanatory variables including gender, age, marital status, nationality, hospital type, and severity level. This approach is based on the assumption that direct costs are independently associated with various clinical and demographic factors.
Using R Package, version 3.6.1 (Foundation for Statistical Computing, Vienna, Austria), univariate and multivariate Ordinary Least Squares (OLS) regression models were carried out to predict Direct Costs, the dependent variable, as a function of the abovementioned demographic and clinical factors. The level of statistical significance was fixed as p<0.05. The dependent variable, Direct Costs, was not normally distributed; therefore, it underwent Box-Cox transformation using a lambda = 0.1, as it was the optimal lambda value for the equation, and one that enabled the greatest Log-Likelihood. The normality of the Box-Cox transformed variable was tested by the creation of a histogram for the residuals, which indicated that the residuals are normally distributed. The OLS model using Box-Cox transformed Direct Costs was tested for other basic assumptions including linearity, homoskedasticity and inter-independence of the explanatory variables. All assumptions were fulfilled, except homoskedasticity that was violated, with Breusch-Pagan test (statistics = 191.93, p<0.001) and NCV test (statistics = 352.34, p<0.001), indicating heteroskedasticity of the model. Consequently, robust regression was carried out using Box-Cox transformed Direct Costs as the dependent variable and the aforementioned demographic and clinical factors as the explanatory variables. Two robust regression models were built. The first model (Robust Model 1) included all six explanatory variables and showed significance for only disease severity and hospital type. The second model (Robust Model 2) included only disease severity and hospital type, and both were significant. Both Robust Models 1 and 2 were tested on the Source Database and their respective mean absolute errors, mean square errors (MSEs), and Root MSEs (RMSEs) were analyzed. These showed that Robust Model 1 had a lower MAE (10466.4) compared with Robust Model 2 (10477.2), indicating better performance of Robust Model 1 in terms of absolute error. However, Robust Model 2 showed lower MSE (271991845.5 versus 278610358.6) and RMSE (16492.2 versus 16691.6), indicating higher accuracy, in terms of error metrics, of Robust Model 2 than Robust Model 1 respectively. Consequently, Robust Model 2 including severity level and hospital type as predictors, was used to predict Box-Cox transformed Direct Costs. The robust regression equation formula with the estimated parameters are presented in the results section. The predictive model was applied on Target database (national data) for each patient, and Estimated Direct Costs (EDC) were calculated by back-transforming Box-Cox Direct Costs using a specific formula (see Results section). Total EDC were calculated for each year (2013-2017) separately, by summation of EDC for all patients in the given year.

Estimation of indirect costs (EIC) following the human capital model
This model is based on the estimation, on the Target Databases (2013-2017), of the disability days costs, which depend on both productivity days lost and GDP per capita. Productivity days loss is divided into days of activity loss (DALS) and productivity years loss (PYLS). Both DALS and PYLS were estimated using a systematic model by Shepard et al. [30] as described by Selck et al. [19]. According to this model, DALS are assumed depending on professional status (schooled or employed), treatment modality (hospitalized versus ambulatory), and severity level (Dengue without warning signs [D.WS-] or with warning signs [D.WS+] versus severe dengue [S.DEN.]); while PALS, which apply in case of disease-related mortality, are calculated as life expectancy minus age at death. For instance, as per the model, an employed individual with D.WS+ loses 6.6 versus 9.9 activity days if he is treated in ambulatory versus hospitalized mode, respectively ( Table 1). Further, the model considers caregiver DALS, assuming the participation of a third person (relatives, friends, etc.) to provide care, assistance, and/or hospital visits to the patient, which results in absenteeism that is estimated as half the DALS of the patient. Specific assumptions were used for estimating DALS of unemployed patients including housewives, jobseekers, retired individuals and under school-age children. Subsequently, indirect costs were calculated per patient by adjusting the GDP per capita for the number of DALS (both patient's and caregivers if applicable); and in case of death, as the sum of per capita GDP per PYLS. Total estimated indirect costs (EIC) of dengue were calculated for each year (2013-2017) using the equation: EIC = patient DALS cost + caregiver DALS cost + PYL cost. Data management, processing and calculations were done using Microsoft Excel 2013 for Windows and SPSS version 21.0 for Windows (SPSS Inc., Chicago, IL, USA).

Mortality No Yes
Treatment modality Ambulatory Hospitalized -

Days of activity loss (DALS)
School days loss 4.

Estimation of total costs (ETC)
Estimated total costs (ETC) were computed as: ETC = EDC+EIC Where, EDC: are the direct costs as estimated using the predictive model EIC: are the indirect costs as estimated using the Human Capital Model Results were presented as the yearly (2013-2017) and total EDC, EIC and ETC. The average cost by patient (ACP) was calculated as: ACP = ETC/N, where N is the number of confirmed cases in the given year. Costs were expressed in the local currency (Saudi Riyal [SAR]) as well as in US$.

Missing data management
Both Source and Target databases were initially managed for missing data by using appropriate imputation methods available on the Statistical Package for Social Sciences version 21.0 for Windows (SPSS Inc., Chicago, IL, USA). Missing data were of two types: missing observations for a given variable, or missing variable (unobserved variable).
1. Missing observations concerned both Source and Target databases and included age, gender, professional status, nationality, severity level and hospital type. These were imputed using multiple regression imputations to predict the missing values for each variable as a function of the other relevant variables (e.g. predicting professional status by age, gender, nationality etc.), and with respect of the local demographic figures (i.e. percentage of students in the general population, etc.). Multiple imputations were ran and analyzed, and those with figures that were inconsistent with the local figures were excluded. For numerical variables like age, where <5% data was missing, mean substitution was used after adjusting for other variables; e.g. for a single non-Saudi male with missing age data, the mean age of single Saudi males in the sample was attributed.
2. Unobserved variables concerned only Target Databases. These included treatment modality and mortality; both variables were unoserved in the national data (Target Databases). Imputation was conducted in two steps. First, each Target Database (2013-2017) was merged individually with the Source Database (N = 717), which was used as a reference dataset for the missing variables. This step was based on the assumption that populations from source (N = 717) and target datasets (2013-2017) are comparable, and thus can be analyzed in a combined dataset. Second, unobserved variables from Target Databases were treated as missing observations and imputed by using multiple or logistic regression imputation, as appropriate, as a function of relevant independent variables from the combined database. Five imputations were performed for each variable and for each combined dataset, resulting in five imputed datasets for each Target Database (2013-2017). Imputed datasets of each year were carefully analyzed, and those which showed the best consistency with the actual epidemiological figures, by reference to source database and clinical sense, were considered valid. Imputations were rejected if they were clearly inconsistent with the Source dataset with respect of the   Table 3.

Building the predictive model for estimated direct costs (EDC)
The first robust regression model, Robust Model 1, which included all six demographic and clinical factors, showed only hospital type and severity level as significant predictors of Box-Cox transformed Direct Costs. The second model, Robust Model 2, was carried out including the two significant variables. Both models are presented in Table 4.
According to the predictive model above, the Box-Cox transformed Direct Costs was estimated using the equation: Where Lambda is the Box-Cox transformation factor = 0.1.

Imputation of mortality and treatment modality
Key figures of the 5 imputations of mortality and treatment modality by severity level for each Target Database (2013-2017) are depicted in Table 5. These include total percentage of hospitalizations, mortality rate, and distributions of both hospitalization and mortality rates by severity level, by reference to source dataset (N = 717). These two variables are relevant for the estimation of EIC using the Human Capital Model. In general, all imputations provided total percentage of hospitalizations that is consistent with Source Database and acceptable ranges of mortality rates by respect of the 95% confidence interval of mortality (0.2%, 1.6%) in source dataset. Inconsistencies lied principally in the distribution of hospitalizations and or mortality rate by severity level, upon which the decision of retention or rejection of the given imputation

PLOS NEGLECTED TROPICAL DISEASES
The economic burden of dengue fever in the Kingdom of Saudi Arabia was based. Note that the high variability of mortality rates in severe dengue in some imputations is due to a small number of cases; thus, imputations with high mortality rates among severe dengue cases (e.g. 70%, 7 out of 10 severe dengue cases) were considered valid.

Estimated direct, indirect and total costs
By processing imputed variables on 2013-2017 national databases, Direct Costs were calculated using the predictive model developed in Phase II analysis and Indirect Costs were calculated using the GDP-based model (Human Capital Model); final results of EDC and EIC with ETC in Saudi Riyal and US$ are presented in Table 6 (Fig 2A), and, outstandingly, PYLS represented the major proportion of EIC (Fig 2B). Thus, PYLS accounted for 63.3% to 83.8% of the estimated total costs, depending on the year.

Correction for under-reporting
Under-reporting was assumed as follows: for every reported dengue case, there are 3 actual dengue cases (i.e., an expansion factor of 3); thus, underreporting was corrected by multiplying by expansion factors (EF3) to scale reported cases. Results of costs after under-reporting correction are presented in Table 7. Thus, the yearly economic burden of dengue fever in Saudi Arabia was estimated as US$110.73 million (result not presented).

Discussion
This is the first study investigating the economic burden of dengue fever infection in Saudi Arabia, based on local data. Its findings emphasized the substantial economic burden of dengue fever in Jeddah and surrounding areas, representing the endemic center of the country and weighing heavily on the national economy. Dengue fever costs were divided into direct and indirect costs, each analyzed using a specific method. For direct costs, we employed a bottom-up approach based on data collected from a representative sample of laboratory-confirmed cases of dengue fever from the highest endemic region in Saudi Arabia during 2016. Direct costs of the sample were modeled as a function of predictors and the predictive model was applied on national data to estimate the respective direct costs. For indirect costs we employed the Human Capital Method, which is an economic model that is used in most of COI studies and is based on the estimation of productivity lost costs by adjusting the GDP per capita [34,40]. In general, three major methods are used to estimate indirect costs in the literature, including the willingness to pay method (WTP), friction cost method (FCM), and human capital method (HCM). In the WTP, the amount which an individual is eager to pay to alleviate the burden of an illness or mortality is quantified. The FCM relies on the productivity lost due to time required to restore the original production level. In such a method, the productivity losses are probably replaceable and thus the main losses are limited to a short period, during which a sick employee could be replaced. FCM is a conservative method because job absenteeism or

PLOS NEGLECTED TROPICAL DISEASES
The economic burden of dengue fever in the Kingdom of Saudi Arabia death are the major determinants of productivity loss. In HCM, the person is viewed as a capital investment and can be valued by his/her societal economic contribution. Therefore, in cases of dengue fever, the temporary productivity loss is estimated as the discounted present value of future earnings expected over the course of the infection, considering the assumption that future earnings are used as a proxy for future productivity [41] .
Although both methods used in this study are statistically and methodologically valid, it is plausible that relying on multiple sources of data would provide more accurate national estimates. Furthermore, besides the results of estimated costs that are discussed in the following sections, data processing sheds light on the considerable lack of economic data with a questionable quality of the available data, which made the cost analysis laborious and yielded some methodological limitations. This denotes the need to integrate effective data collection and management procedures as part of disease surveillance and control strategies.

Underestimation: A problematic aspect
For the total 15,369 patients included over the 5 years in the current study, the estimated yearly cost of dengue illness was US$110.17 million after assuming an EF of 3. Considering US dollar inflation rates in 2018 [42], our annual estimate was relatively in agreement with that of a recent systematic review that investigated the global economic burden showing an annual economic burden of US$99.25 million [21]. However, our figure was higher than the majority of other reports. A recent multinational study in Southeast Asia comprising 12 countries has reported an annual cost of US$31. study are explained by the GDP per capita in Saudi Arabia being much higher than many other countries. Additionally, our estimates showed that a substantial proportion of indirect costs was related to productivity years loss, which is directly associated with the country's GDP. Furthermore, the estimation of productivity years loss is directly impacted by the relatively high life expectancy and the young population of Saudi Arabia. Thus, productivity years loss accounted for 63 to 82% of the total costs in the present study. The productivity loss contributes to the largest proportion of total costs in several other COI studies, accounting for 50-60% of the total costs [26,30,43], which is in line with our findings. On the other hand, vector control measures, which are considered as indirect costs, were reported to represent 40-72% of costs implied on healthcare systems when estimated in different studies [34,47,48]. Regarding the EF, the value used in the present study is relatively consistent with an EF of 3.79 in a Malaysian investigation that relied on merged data from the literature, expert opinion, and implementing a Delphi process [36]. However, our EF is significantly lower than that from studies published in Southeast Asia, Thailand and Cambodia (EF = 7.8-9.1) and Puerto Rico (EF = 10)[37-39].
Therefore, it is acceptable that our ETC estimate and EF are conservative, probably considering three major factors. First, disease estimates in the current study reflect the patients with a symptomatic dengue in its acute phase, which potentially affects the quality of life (QOL) of patients; however, chronic fatigue can occur post-dengue infection, especially in females, elders and patients having acute phase chills and may persist from 2 months to 2 years in 8.5%-57% of the patients [49][50][51][52]. These chronic consequences and their impact on the QOL may ultimately affect the total economic burden. However, they were not considered in the present study and hence the costs are likely to be underestimated. Second, economic calculations in the present study did not entail the costs of preventive measures, vector control, or surveillance programs, which may increase the overall costs by 20% to 43% [26,53,54]. Third, some additional costs can presumably exist, such as intangible costs on the healthcare system, the impact of dengue and its periodic epidemics in Saudi Arabia on tourism and other economic sectors that could not be elucidated in our study. These might be of a notable effect that renders the net cost underestimated.

The potential reasons of cost variation
The variations in cost estimation may be affected by several factors. First, the discrepancy of costs between hospital types as indicated in the current study. More specifically, hospital type had a significant effect on the cumulative changes in EDC over the study period and was a significant predictor of direct costs of dengue. Similarly, in Thailand, there was a significant difference in the COI between large referral centers and community or provisional hospitals [55] . The degrees and magnitudes of variation of COI should be evaluated in multiple facilities of the same type as suggested previously [56]. Second, transportation to healthcare facilities may contribute to outpatient costs [55,57]. However, the impact of transportation barriers seems to be only apparent in earlier studies in Saudi Arabia as revealed by 39% of patients [58], while recent evidence indicated no association between the distance to healthcare facilities and patient's satisfaction [59].

The cost-effectiveness of vaccination and other control programs
Dengue vaccination and dengue control programs may offer a cost-effective approach, with a significant disparity among developing and developed countries. In the early 1990s, Shepard and Halstead [60] found that dengue vaccine was cost-effective by saving an average of US $1,440 per disability-adjusted life year (DALY) and US$92.4/death averted in developing countries with poorly-developed healthcare systems. However, 10 other Southeast Asian country-based analyses [61] assumed vaccination costs of $10/dose in the private sector and US $0.50/dose in the public sector with an administration cost of US$3.50/dose. More recently, Zeng et al. [62]concluded that dengue vaccine has the potential to reduce direct and indirect costs per capita by 22% and 23% in Latin America and Asia, respectively. However, the work by Zeng et al. did not consider serological status prior to vaccination, which is highly recommended due to risk of severe dengue among individuals without prior exposure to the virus. Therefore, other researchers assessed the cost-effectiveness of combined serological screening and vaccination strategy, where only seropositive people would be candidate for vaccination [63][64][65]. Such strategy aimed at reducing DENV transmission and the related economic burden. These authors analyzed different scenarios that considered the variability, across settings, of DENV transmission rate, diagnostic performance of the serological screening test (i.e. sensitivity and specificity), the cost of vaccination and serotesting, and the value of disabilityadjusted life-year. Although mitigated, findings generally indicated that cost-effectiveness of such campaigns could be expected in settings with a high transmission rate of DENV and high burden of dengue, provided that the sensitivity of the serotesting is adequate. On the other hand, a high transmission rate may downplay the added value of serotesting in such a strategy. Furthermore, one of the aforementioned models suggests that repeated serotesting would improve health benefits and may result in significant economic savings, using a return-oninvestment (ROI) model, thus achieving cost-effectiveness [65].
In light of these findings, it is imperative to assess the efficacy and cost-effectiveness of dengue vaccination campaigns in Saudi Arabia, besides other preventive and surveillance programs, using local epidemiological and economic indicators, with the aim of reducing the substantial health and economic burden of the disease [63][64][65].

Limitations
In this study, we evidenced some limitations that might affect the interpretation of the outcomes. Inclusion of patients in a hospital setting indicates that they experienced relatively moderate to severe symptoms and, therefore, a proportion of mildly symptomatic patients may have remained undiagnosed and escaped inclusion. Thus, despite the correction factor (EF = 3), the included sample might be unrepresentative of the whole population. Furthermore, the used EF was based on existing literature rather than real evidence from Saudi Arabia. Applying a constant EF from non-local studies would possibly be problematic given that EF may differ according to the intensity of transmission [66], transmission season [67], patients' age [39], disease severity [68], level of access to healthcare [38], and local variations [69]. From another perspective, it is possible that the EF may be inflated relative to the actual instances when a laboratory-confirmed patient is referred from a private hospital to the public sector after a positive result has been obtained and hence may be reported twice. Moreover, the EFs of the private sector are uncertainly estimated, particularly for ambulatory cases.
The method used for indirect costs estimation is limited by the assumptions regarding DALS due to absenteeism, which may be over-estimated especially in non-severe cases. On the other hand, this method was limited by the multiple imputations required to correct missing variables. This urges the implementation of reliable data collection and management systems to enable more accurate estimations.
Drivers of higher costs, such as prolonged disabling symptoms and effects on tourism and other economic sectors that might be associated with dengue epidemics were not considered in our statistical analysis. In addition, the inability to include costs of vaccination and vector control strategies (as they were beyond the objectives of the study) would eventually add to the limiting factors.

Future implications
Evidently, there is a need to conduct future local studies, on a larger scale, that entail a multiregional methodological approach to provide robust national evidence. This is to overcome the potential bias that may emerge from estimating the economic burden in an endemic region like Jeddah, along with the existence of substantial spatial differences across the Kingdom.
In general, future studies should be based on standardized monetary units to facilitate the comparisons between countries of different purchasing powers and the outcomes should be analyzed considering the financial structure of local healthcare systems. It is also recommended to combine medical costs with costs related to reduced QOL, tourism estimates, and vector control expenses. A comprehensive analysis of the economic aspects in the private sector is needed to help guide decision-making and policy-making, particularly in terms of employing vaccine campaigns and the development of antiviral drugs. Estimates of productivity loss could be derived via the human capital approach rather than self-reported information that can be subject to recall bias. However, more accurate assumptions are needed for cost estimations for those working in both the formal and informal sectors. Another crucial element that should be taken in mind is quantifying the cyclical variations to capture the common patterns of disease transmission to reveal accurate cost estimates during the endemic and epidemic times. This could be partially done via collecting data over multiple years.
Importantly, intensified vector control should be a critical component of health sector planning in Saudi Arabia, particularly in light of mass gatherings during Hajj and Umrah, primarily constituted from populations from Southeast and South Asia, and the possible transmission of dengue and other vector-borne neglected tropical diseases [70][71]. The nature and duration of disease burden and its economic consequences should be studied following these religious rites.
Finally, it is essential to establish future comparative analyses of the cost-effectiveness of a dengue vaccine. This should include avoiding the inherent difficulties implied by such studies as per experts' recommendations, including the application of well-defined and cited assumptions, considering the vetted costs of vaccination and its possible side effects as well as surveillance programs aiming to monitor the safety of new vaccines, utilizing natural units to report cost-effectiveness (such as DALY, death averted, admission averted, etc.), and the estimated effects on mortality since policymakers may wrongfully expect a high mortality.

Conclusions
Rapid urbanization, global trade, and the exceptionally great numbers of worldwide visitors during Hajj and Umrah have all placed the Kingdom of Saudi Arabia at a significant risk of introduction of several vector-borne tropical diseases. Of them, dengue fever has been established as an endemic disease for two decades and this could be emphasized in the current study. Dengue has a substantial local economic burden that costs US$110.17 million per year resulting in an average cost of US$11 947.6 per patient. Costs were highly disparate across the different types of healthcare institutions and productivity years loss accounted for approximately 80% of the total costs.
These estimates, as with other global estimates, are subject to underestimation given several factors, including the use of unstandardized expansion factors, likelihood of exclusion of the mildly-affected cases, underestimation of other symptoms and chronic complications that have a prolonged economic impact, and exclusion of vector control and other preventive activities from the analysis. In light of study limitations, there is a need to conduct future studies employing a multiregional approach, using standardized monetary units¸quantifying cyclical patterns of the vector and health burden, and investigating the cost-effectiveness of a dengue vaccine.