The health and economic impact and cost effectiveness of interventions for the prevention and control of overweight and obesity in Kenya: a stakeholder engaged modelling study

Abstract

Background

The global increase in mean body mass index has resulted in a substantial increase of non-communicable diseases (NCDs), including in many low- and middle-income countries such as Kenya. This paper assesses four interventions for the prevention and control of overweight and obesity in Kenya to determine their potential health and economic impact and cost effectiveness.

Methods

We reviewed the literature to identify evidence of effect, determine the intervention costs, disease costs and total healthcare costs. We used a proportional multistate life table model to quantify the potential impacts on health conditions and healthcare costs, modelling the 2019 Kenya population over their remaining lifetime. Considering a health system perspective, two interventions were assessed for cost-effectiveness. In addition, we used the Human Capital Approach to estimate productivity gains.

Results

Over the lifetime of the 2019 population, impacts were estimated at 203,266 health-adjusted life years (HALYs) (95% uncertainty interval [UI] 163,752 − 249,621) for a 20% tax on sugar-sweetened beverages, 151,718 HALYs (95% UI 55,257 − 250,412) for mandatory kilojoule menu labelling, 3.7 million HALYs (95% UI 2,661,365–4,789,915) for a change in consumption levels related to supermarket food purchase patterns and 13.1 million HALYs (95% UI 11,404,317 − 15,152,341) for a change in national consumption back to the 1975 average levels of energy intake. This translates to 4, 3, 73 and 261 HALYs per 1,000 persons. Lifetime healthcare cost savings were approximately United States Dollar (USD) 0.14 billion (USD 3 per capita), USD 0.08 billion (USD 2 per capita), USD 1.9 billion (USD 38 per capita) and USD 6.2 billion (USD 124 per capita), respectively. Lifetime productivity gains were approximately USD 1.8 billion, USD 1.2 billion, USD 28 billion and USD 92 billion. Both the 20% tax on sugar sweetened beverages and the mandatory kilojoule menu labelling were assessed for cost effectiveness and found dominant (health promoting and cost-saving).

Conclusion

All interventions evaluated yielded substantive health gains and economic benefits and should be considered for implementation in Kenya.

Background

Globally, non-communicable diseases (NCDs) are the leading cause of deaths (74% of global deaths) and morbidity (64% of disability adjusted life years [DALYs]) [1, 2]. Overweight (Body mass index [BMI] of 25.0-29.9 kg/m²) and obesity (BMI ≥ 30.0 kg/m²) have been identified as leading risk factors for NCDs [3,4,5]. The global increase in mean BMI [6, 7] has contributed to a substantial increase of this NCD burden, including in many low- and middle-income countries (LMICs) countries such as Kenya [2, 8, 9]. An estimated 27% of the adult population in Kenya has overweight or obesity (38.5% women [~ 5 million] and 17.5% men [~ 2.2 million]) [10]. The increased BMI related NCD burden greatly impacts individuals’ economic livelihoods and strains the country’s health care system that is still battling communicable, maternal, neonatal and other nutritional diseases, which although in decline, still dominate Kenya’s disease burden [1]. If the current trends continue unabated, it will lead to increased rise in the NCD burden putting further strain on the health system. In our recent study, we found that over the lifetime of the 2019 Kenyan population, high BMI could cause losses of approximately 83.5 million health-adjusted life years (HALYs) (~ 1.7 HALYs per person) and decrease health-adjusted life expectancy by 2.3 years for females and 1.0 years for males [11]. The magnitude of the avoidable high BMI-related disease burden underscores the need to prioritise the control and prevention of overweight and obesity.

In Kenya and other Sub-Saharan Africa countries, high BMI has been linked to changes in dietary patterns and nutrient intakes [12, 13]. These changes are fuelled by factors such as urbanisation, increased income, and changes in the food systems that have seen the expansion of transnational food and drink corporations into ‘emerging markets’ [12,13,14,15]. The changes are characterised by a departure from indigenous foods that are often high in carbohydrate, fibre and low in fat and sugar. The indigenous foods are replaced with ultra-processed products, foods high in saturated fat, sugar and salt, low in fibre and other key nutrients [12, 13, 17]. In order to meet global and national obesity reduction targets [17,18,19], creating healthy environments that stimulate physical activity and encourage a healthy diet may be more impactful than targeting individual behaviour [20, 21]. Policy options directed at the ‘obesogenic’ environment complement education campaigns and social marketing [22]. Evaluation of such policies or preventive interventions enables the identification of those that are impactful and cost effective [23, 24]. This informs judgments on the allocation of resources and priority setting for health. In public health, measurement of outcomes is challenging since most health effects of behaviour change occur only after many years. As a practical alternative to direct measurement, estimates of health outcomes can be obtained through epidemiological modelling [23, 25].

In this study, we assessed interventions for the prevention and control of overweight and obesity in Kenya to determine their potential health and economic impact and cost effectiveness. Intervention selection was informed by a stakeholder engagement process with policy makers in Kenya [26]. To our knowledge, no studies have assessed the potential economic and health impact and cost effectiveness of interventions that aim to prevent and control overweight and obesity in adults in Kenya.

Methods

Overview

In this study, we applied the assessing cost-effectiveness (ACE) approach which has been developed as a priority setting tool that enables evidence-based decision making [27, 28]. Stakeholder engagement was part of the due process and this has been reported elsewhere [26]. We used a lifetime horizon in our assessment and targeted the 2019 Kenya population modelled over their remaining lifetime. We explicitly modelled 37 high BMI related NCDs. We used the WHO recommended generalised cost-effectiveness analysis approach where we compared interventions against a ‘do nothing’ scenario [29]. With respect to body mass, in Kenya, the ‘do nothing scenario’ may also closely reflect the ‘current practice’ in the model 2019 base year. This is because most of the government policy actions on prevention of overweight and obesity are still at the development stage [30]. This comparator scenario is modelled as 2019 BMI levels in Kenya with a linear upward trend continued unabated for 25 years. Baseline BMI levels are taken from published national survey results [10, 31]. We derived the BMI trend from age- and sex-specific mean BMI data for Kenya from 1975 to 2016 as provided in the NCD Risk Factor Collaboration (NCD-RisC) study whose primary data sources are Kenya national surveys done over the years [7]. We checked the BMI levels from the NCD- RisC study (mean and lower, upper uncertainty intervals) against the measured mean (and standard deviation) BMI levels in Kenya from 1993 to 2015 national survey data [31,32,33,34,35,36] and found the data comparable. In Supplementary File (SF) Fig. 1 we illustrate this using the 2015 BMI data from the two sources [7, 31]. We fitted both the linear and second order polynomial equations trends to the NCD-RisC data and found the results comparable with the projected rise in BMI being slightly lower when the linear trend was applied. The linear trend equations were subsequently used to derive the specific mean BMI for our base year 2019 from 2015 Kenya ‘STEPwise approach to Surveillance of NCD risk factors’ (STEPS) survey data and to predict future mean BMI. Consistent with shifts observed in the work done by Fogel [37], the standard deviation was considered to shift in equal proportion to the mean (i.e., a ratio of 1:1 is assumed). As an example, in SF Fig. 2, we present the weighted average BMI levels for the 20-24-year-old age group in future years.

Our study adhered to the Consolidated Health Economic Evaluation Reporting Standards 2022 (CHEERS 2022) statement: updated reporting guidance for health economic evaluations [38].

Intervention selection

Our selection of interventions for inclusion in this study was drawn from broad strategies identified in a stakeholder engagement process with policy makers in Kenya [26]. We asked the stakeholders to identify existing and new strategies they considered relevant and appropriate for the prevention and control of overweight and obesity in Kenya. From that list of broad strategies, each stakeholder selected their top three for inclusion in our ACE modelling study. We calculated total weighted scores and selected the two highest ranked strategies for inclusion in our study [26]. These were broad strategy scenarios which included: 1) a research-based strategy where the Kenya Agricultural & Livestock Research Organisation (KALRO) promotes and coordinates research to investigate the nutritional value of indigenous foods and 2) a health promotion and education strategy that does not only tell people to eat healthy diets but one that also creates a healthy environment [26]. Under the first strategy scenario, it is envisioned that the research-based evidence would then be used to promote consumption of indigenous foods found to have high nutritional value. For the second strategy scenario, we considered interventions that create a healthy environment as those that seek to reverse the obesogenic nature of environments by influencing environmental factors such as structures, systems, laws, policies, and sociocultural norms [39, 22]. Next, we searched literature to define what the proposed broad strategies might look like in practical terms, describe them as interventions that can be modelled and identify evidence of intervention effectiveness.

Definition of interventions and evidence of effect

We considered the feasibility of the two broad strategies proposed, timelines required for implementation of specific scenarios/interventions and achievement of effect and, availability of evidence on intervention effect. Because it is hard to fully describe what the broad strategies would look like in future, estimate their implementation costs, and define effect on body mass, we focused on examples of feasible interventions that Kenya could implement as part of initial steps of the broad strategies proposed by stakeholders. We modelled the first broad strategy as two possible intervention scenarios that increased consumption of healthy indigenous foods: change in consumption levels related to supermarket food purchase and change in national consumption levels back to the 1975 average levels of intake (Fig. 1). Under the second broad strategy we selected two policy interventions that Kenya could implement towards creation of healthy food environment: mandatory kilojoule labelling on food served in formal sector restaurants and a tax on sugar-sweetened beverages (SSBs).

Fig. 1

Mapping out implementation scenarios for the selected interventions

*Supermarket purchase is seen to contribute to a shift from indigenous foods to processed foods and drinks [40, 41]

Where the effect of an intervention was reported as changes in calories, we multiplied the calories by 4.18 to convert them to kilojoules [42]. We calculated resultant body weight changes using previous work that has determined that in adults, a change in net energy intake of 94 kJ per day (95% CI 88.2 to 99.8) corresponds to 1 kg change in body weight [43,44,45]. Meaning that in our model, a reduction in daily energy consumption of 94 kJ was assumed to cause 1 kg in weight loss. We then converted weight changes to changes in BMI using the average height by age and sex derived from the Kenya STEPS survey (SF Table 1) [10]. All interventions were modelled as population-based interventions. For comparability, we assumed that the estimated BMI changes resulting from the four interventions were maintained over the lifetime of the modelled population. In Table 1, we describe how each intervention was defined for modelling and the supporting evidence of effect from literature. Additional details of this process are presented in the SF. For each modelled intervention, we present the age and sex specific effect size expressed as a change in BMI units in SF Table 2.

Table 1 Summary of how each intervention was modelled and evidence of effect from literature

Estimation of intervention costs

We based our estimates on the World Health Organization (WHO) NCD costing tool country specific estimates for Kenya [63]. This tool uses an ingredients-based (bottom up) approach where all resources required to deliver an intervention are identified, quantified, and a price is assigned to each resource. For the strategy on increased consumption of healthy indigenous foods following investment in research, we identified additional cost estimates from the Kenya National food and nutrition security policy implementation framework 2017–2022 [64]. From the NCD costing, we draw estimates for the cost of the intervention activity ‘Increasing consumption of healthy indigenous foods through marketing’ (SF Table 10). However, we were not able to cost all aspects of our first idealistic intervention. This is because there are many possible approaches to this broad strategy proposed by the stakeholders. For instance, a realistic two-year research plan can include literature analysis of these indigenous foods or food composition analysis to make inferences. Another possibility is implementation of a large research program within a food or agriculture research organisation where costs are a portion of the organisation’s annual budget. This scenario would require a substantial program and sustained effort over time to get the results. It is difficult to know what it would take to get the results or evidence and hence describe exactly what that program would look like. We report our partial costing process and estimates for this broad strategy in SF Table 10 to facilitate further work in this area. The two intervention scenarios under this first broad strategy were excluded from the cost effectiveness analysis, but we calculate the maximum intervention cost that could be incurred whilst ensuring that intervention remains cost-effective and, in the results, we provide this value as a maximum justifiable cost estimate.

For costing the two specific interventions on SSBs tax and menu kilojoule labelling on food served in formal sector restaurants, we followed the costing examples for the six cost components needed for implementation of similar upstream regulatory policy: human resources (for program management, promotion and media advocacy, law enforcement and inspection and national-level technical assistance); training for program staff; meetings involving external agencies; mass media; supplies and equipment [63]. In the absence of a primary costing study, the NCD costing tool provided the best estimates for the items costed under each of the six cost components. Intervention costs were applied for 100 years to reflect the lifetime of the population with a constant implementation cost applied from year three onward (Table 2). We inflated all costs to 2019 (our model base year) based on Kenya consumer price indices [65] and converted the Kenya shillings, to US dollars using the world Bank’s official 2019 exchange rate of 102 [66]. In the model, we applied discounting to future intervention costs for the lifetime period (3% discounting for the base case).

Table 2 Costs estimates for the policy interventions included in the cost effectiveness analysis*

Estimation of healthcare costs

We describe the estimation of the total and disease specific healthcare costs in Kenya in our previous work [67]. In summary, the 2019 total healthcare costs for Kenya were derived from the 2020 WHO Global Health Expenditure Database (SF Table 11) [68]. We apportioned the total healthcare costs across specific sex and age groups based on age specific annual per capita health spending and sex specific spending on healthcare in Kenya [69]. We used data from five studies to calculate estimates of costs per incident or prevalent case of modelled diseases (SF Table 12) [70,71,72,73,74]. Of the specific diseases modelled, we did not identify any literature with costs for eight, i.e. low back pain, osteoarthritis hip, osteoarthritis knee, gout, Alzheimer’s disease and other dementias, cataract, gallbladder and biliary diseases and atrial fibrillation and flutter. For chronic kidney disease, only costs of kidney transplants and dialysis were identified. Using national survey data, we adjusted the disease costs to account for the percentage of people unwell who did not seek care hence did not incur healthcare costs [69].

To derive the yearly costs of all other diseases per person, we subtracted the age and sex specific costs of all modelled diseases from the total health expenditure in the respective sex and age groups (SF Table 11). These costs are included because as preventive interventions prolong life, additional health expenditure is expected in those added years of life [75]. We modelled the healthcare costs in United States Dollar (USD) as extracted from the primary sources and database [68, 70,71,72,73,74]. We attempted to quantify uncertainty by setting a normal distribution for health care costs assuming the standard deviation to be 20% of the point estimates.

Proportional multi-state lifetable modelling

We developed the Kenya Obesity Model, a proportional multi-state lifetable (pMSLT) model [76] to quantify the health and economic impact of changes in prevalence of high BMI from selected preventive interventions for the 2019 Kenya population (50.2 million) (SF Table 13). We modelled the 2019 population from age zero years to avoid the missing cohort effect, meaning that we capture the impact of the interventions on the younger age groups once they are 20 years and more into the future. Details of the Kenya Obesity Model are published in our previous works [11, 67]. The pMSLT model has been used in previous evaluation studies to estimate health and economic impacts and cost effectiveness of various obesity-related preventive strategies [20, 59, 77, 78]. The model is divided into a general section, that is a standard cause elimination life table (main lifetable), and a separate section for each disease with an independent illness-death (Markov) process [76]. The model simulates the 2019 Kenya population (≥ 20 years) and estimates intervention related changes in BMI on the incidence of the 37 modelled high BMI related disease types over the lifetime of the population. The modelled diseases are: type 2 diabetes, six cardiovascular conditions (atrial fibrillation and flutter, hypertensive heart disease, ischaemic heart disease, ischaemic stroke, intracerebral haemorrhage, subarachnoid haemorrhage), 18 cancers (breast [female], colon and rectum, gallbladder and biliary tract, kidney, leukaemia [five types], liver [three types], multiple myeloma, oesophageal, ovarian, pancreatic, thyroid, uterine), chronic kidney diseases (four types), four musculoskeletal diseases (gout, low back pain, osteoarthritis hip and osteoarthritis knee), four diseases categorised as ‘other diseases’ for ease in reporting results (Alzheimer’s disease and other dementias, asthma, cataract, gallbladder and biliary diseases).

In comparison with a ‘do nothing’ scenario where the current BMI levels and trend continue unabated for 25 years, reduced incidence of diseases in the intervention population overtime results in reductions in prevalence and mortality (apart from musculoskeletal diseases and cataract that are not linked to disease specific mortality in the model). The changes in incidence after an intervention related change in BMI were calculated using the potential impact fraction (PIF) applying the distribution shift method [79]. The PIF is a measure of effect that calculates the proportional change in average disease incidence (or prevalence or mortality) after a change in the exposure of a related risk factor [79]. The distribution shift method assumes a continuous risk-factor (BMI) distribution (modelled as lognormally distributed [11]) with continuous relative risks (RRs) from GBD 2019 [80] modelled as normally distributed [81]. The RR estimates (mean, lower and upper levels) by age and sex are the relative risk of morbidity (incidence in our model) from a high-BMI-related disease, per 5 BMI-unit (5 kg/m²) increase. We consider the distribution of BMI within each age-sex category and only model benefits for the part of the distribution above BMI 22·5 kg/m². In the PIF calculation, we used 22·5 kg/m² as our lower boundary BMI. This is in line with the available evidence which suggests that the risk of NCDs that are commonly seen in adults starts rising from around BMI 22·5 kg/m² [80]. We applied the following PIF equation:

$$PIF = \frac{{\int\limits_a^b {RR(x)P(x)dx\, - \int\limits_a^b {RR(x){P^*}(x)dx} } \,}}{{\int\limits_a^b {RR(x)P(x)dx} }}$$

x: risk factor level, RR(x): relative risk function, P(x): original risk factor distribution, P*(x): intervention risk factor distribution, a,b: the integration boundaries, dx: integration with respect to x

The changes in incidence of the disease subsequently lead to changes in morbidity and mortality rates. Changes in disease-related quality of life at every age were calculated using disease-specific disability weights [2, 11, 82]. The new disease specific mortality and morbidity rates and changes in costs from the disease sections are fed into a life table to calculate the number of HALYs and total cost savings. Changes in healthcare expenditure were estimated both for modelled high BMI related NCDs and overall healthcare costs in added years of life [75]. A detailed report on the sources of data, preparation of data and estimates of the epidemiological input data used in the model is provided elsewhere in our previous work [11].

Cost-effectiveness modelling

Considering a health system perspective, we conducted a cost effectiveness analysis for two interventions, the 20% SSB tax and mandatory kilojoule menu labelling. We calculated the incremental cost-effectiveness ratios (ICERs), defined as the difference in net costs of the intervention compared to no intervention comparator, divided by the difference in HALYs. Net costs included intervention costs and healthcare costs (including costs in added life years). We assumed all interventions to be fully implemented and running at their full effectiveness potential. We assumed that the impact of interventions on BMI (intervention effect) were maintained over the lifetime of the modelled population.

Given the absence of cost effectiveness thresholds in Kenya, we used the WHO benchmarks for definition of cost-effectiveness for each modelled intervention. We defined a very cost-effective intervention as ICER < 1,801 USD (i.e., the gross domestic product [GDP] per capita for Kenya in 2019) [83] per HALY gained, and a cost-effective intervention being defined as ICER < 5,403 USD (i.e., 3 times the Kenya GDP per capita) per HALY gained. Cost-saving was defined as having a negative net cost.

Other outcomes reported

We computed the tax revenue that would be generated following implementation of the SSB tax. The assessment of productivity gains was also included as this was already built into the model in our previous work [67]. Productivity gains were not part of the cost effectiveness analysis. We used the Human Capital Approach [24] to estimate productivity gains resulting from a reduction in high BMI -related 1) mortality, 2) mortality and morbidity (combined) and 3) morbidity. Further details on this are provided in the SF. Additionally, we included net monetary benefit accruing from the two interventions assessed at this stage. This was calculated as the total healthcare cost savings and productivity gains realised from each intervention less total intervention cost. The productivity gain estimate used in the net monetary benefit calculation is the obesity-related mortality and morbidity (combined) outcome [67]. We also report the net monetary benefit without inclusion of productivity gains. In the main analysis, we applied a discount rate of 3% to costs and HALYs as recommended in Drummond et al [24]. We shared details on costing process, cost estimates, cost effectiveness threshold for Kenya with the stakeholders for their review and comment.

Uncertainty and sensitivity analysis

While incorporating uncertainty from model inputs, we conducted a probabilistic sensitivity analysis using Monte Carlo simulation with bootstrapping (2000 iterations) to estimate the 95% uncertainty intervals around BMI, HALYs and cost outcomes in the base case scenarios.

This was implemented using Ersatz version 1.35 software [84]. The point estimate and 95% uncertainty intervals (UI) were defined as the 50th and 2.5th -97.5th percentiles, respectively.

We conducted univariate sensitivity analyses to explore the impact of variation in discount rates (0% and 5%). For the SSB tax intervention, additional sensitivity scenarios were included by varying the tax from 20% (base case) to 10% and 30%, varying the pass on rate from 100% (base case) to 80% and 120% and, by applying cross price elasticity to the base case scenario. In order to compare like for like in the sensitivity analysis we run all scenarios with uncertainty off.

Ethics approval was not required for this analysis. However, this study was carried out as part of a larger study that has ethical approval from the Griffith University Human Research Ethics Committee (GU Ref No: 2019/707) and the Kenyatta National Hospital/University of Nairobi Ethics & Research Committee (P81/02/2021).

Results

Reduction in overweight and obesity

If implemented in 2019, a 20% tax on sugar sweetened beverages could result in a reduction of ~ 44,232 people with overweight (Males = 28,902 UIs 22,229 to 37,442, Females = 15,330 UIs 11,800 to 20,052) and ~ 32,140 people with obesity (Males = 10,090 UIs 7,820 to 12,802, Females = 22,050 UIs 17,504 to 27,497) (Fig. 2). The mandatory kilojoule menu labelling could result in a reduction of ~ 33,691 people with overweight (males = 20,665 UIs 7,220 to 34,126, Females = 13,026 UIs 4,531 to 21,578) and ~ 27,076 people with obesity (Males = 7,579 UIs 2,655 to 12,492, Female = 19497 UIs 6,817 to 32,180) while the change in consumption levels related to supermarket food purchase intervention could result in a reduction of ~ 1.1 million people with overweight (Males = 634,968 UIs 474,251 to 788,201, Females = 416,069 UIs 296,171 to 538,257) and ~ 731,268 people with obesity (Males = 210,734 UIs 161,637 to 255,861, Females = 520,534 UIs 389,434 to 645,705). The change in national consumption levels back to the 1975 average levels of energy intake could lead to the highest gains with a reduction of over 4.3 million people with overweight (Male = 1,955,479 UIs 1,900,480 to 2,009,639, Female = 2,378,665 UIs 2,250,201 to 2,510,957) and over 2.3 million with obesity (Males = 534,684 UIs 524,115 to 544,838, Females = 1,771,755 UIs 1,722,546 to 1,820,027). Apart from the supermarket food purchase scenario where the evidence of effect was a change in BMI, the intervention effect was modelled at the level of change in consumption.

Fig. 2

Reduction in number of people with overweight or obesity

kJ: kilojoule, BMI: body mass index, SSB: sugar sweetened beverage. *Change in national consumption levels: Change in national consumption levels back to the 1975 average levels of energy intake

Changes in disease incidence, prevalence, and mortality

Over the first 25 years, implementation of a 20% tax on sugar sweetened beverages could prevent an estimated total of 80,060 incident cases of diseases across the diseases modelled. The results tables indicate the uncertainty intervals for all estimates reported. Over the same period, a total of 5,083 deaths for type 2 diabetes (T2DM), cardiovascular diseases, chronic kidney disease (CKD), cancers, Alzheimer’s disease and other dementias, asthma, gallbladder and biliary diseases could be avoided (Table 3). The greatest impact of averted new disease cases was seen in T2DM. In the 25th year (the year 2044), across all modelled diseases, ~ 42,511 prevalent cases of disease could be avoided.

The mandatory kilojoule menu labelling intervention could see a total of 57,319 incident cases of diseases avoided over the first 25 years. In the same period, a total of 4,157 deaths avoided for T2DM, cardiovascular diseases, CKD, cancers, Alzheimer’s disease and other dementias, asthma, gallbladder and biliary diseases (Table 3). In the 25th year, ~ 28,389 prevalent cases of disease could be avoided. The greatest impact of averted new diseases cases was seen in musculoskeletal diseases followed by T2DM, cardiovascular conditions and CKDs (Table 3). For the other two interventions where a change in national consumption levels back to the 1975 average levels of energy intake and in consumption levels related to supermarket food purchase were modelled, the greatest reductions of new disease cases were also seen in musculoskeletal diseases followed by T2DM, cardiovascular conditions and CKDs.

For the change in consumption levels related to supermarket food purchase, over the first 25 years, a total of 1.2 million incident cases of diseases could be averted and a total of 72,581 deaths avoided for T2DM, cardiovascular diseases, CKDs, cancers, Alzheimer’s disease and other dementias, asthma, gallbladder and biliary diseases. In the 25th year (year 2044), ~ 699,764 prevalent cases of modelled diseases could be avoided (Table 3). A change in national consumption levels back to the 1975 average levels of energy intake yielded the largest magnitude of impact. Over 4.3 million incident cases of diseases could be averted, and 247,594 deaths avoided over the first 25 years (Table 3). In the year 2044, ~ 2.4 million prevalent cases of disease could be avoided. For each intervention, changes by disease are reported in SF Tables 14 to 17.

Table 3 Projected number of incident cases, prevalent cases and avoided deaths by disease group

Changes in health adjusted life years, life years and overall deaths

Over the lifetime of the 2019 population, the impact of the 20% tax on sugar sweetened beverages could translate to ~ 203,266 HALYs, ~ 151,718 HALYs for the mandatory kilojoule menu labelling, over 3.6 million HALYs for the change in consumption levels related to supermarket food purchase intervention and over 13 million for the change in national consumption levels back to the 1975 average levels of energy intake (Table 4). In the year 2044, over 3,000 deaths could be avoided following the implementation of a 20% tax on sugar sweetened beverages. For the same period, implementation of the mandatory kilojoule menu labelling could see over 2,000 deaths avoided, a change in consumption levels related to supermarket food purchase could lead to over 53,000 deaths avoided while a change in national consumption levels back to the 1975 average levels of energy intake could see ~ 185,869 deaths avoided.

Table 4 Effects of each intervention on HALYs, LYs and overall deaths

Economic impact and cost effectiveness

Over the lifetime of the 2019 population, the 20% SSB tax could lead to a reduction in healthcare cost by about USD 140 million and the mandatory kilojoule menu labelling could result to healthcare cost savings of USD 83 million (Table 5). A change in consumption levels related to supermarket food purchase could yield over USD 1.9 billion in healthcare cost savings and a change in national consumption levels back to the 1975 average levels of energy intake could yield over USD 6.2 billion (Table 5). Each intervention resulted in large productivity gains. Over the lifetime, productivity gains due to obesity-related mortality and morbidity (combined) were in excess of USD 1.8 billion for the 20% tax on sugar sweetened beverages, over USD 1.2 billion for the mandatory kilojoule menu labelling, over USD 27 billion for the intervention on a change in consumption levels related to supermarket food purchase and, over USD 92 billion for the change in national consumption levels intervention.

The intervention cost (discounted) for the 20% tax on sugar sweetened beverages was estimated to cost USD 9.9 million while the mandatory kilojoule menu labelling would cost USD 12.0 million (Table 5). These two interventions were found to be very cost effective when cost offsets were not included: USD 49 per HALY gained (95% UI 39–60) for the 20% tax on sugar sweetened beverages and USD 92 per HALY gained (95% UI 47–202) for the mandatory kilojoule menu labelling. When cost offsets were included the two interventions were dominant (health promoting and cost-saving) (Table 5). For 20% tax on sugar sweetened beverages, the net monetary benefit without productivity was ~ USD 0.13 billion and ~ USD 2 billion with productivity. The mandatory kilojoule menu labelling resulted to a net monetary benefit of ~ USD 0.07 billion without productivity and ~ USD 1.3 billion with productivity. In addition to the above economic gains, the 20% tax on sugar sweetened beverages would also raise USD 71 billion in revenue for the government or USD 98 billion without discounting (Table 5).

Table 5 The economic impact and cost effectiveness

Total refers to estimates for both male and female. kJ: kilojoule, SSB: sugar sweetened beverages, UI: Uncertainty interval, USD: US dollars. ICER: incremental cost-effectiveness ratios, dominant (dominant: health promoting and cost-saving).

The following modelled diseases were not costed: low back pain, osteoarthritis hip, osteoarthritis knee, gout, Alzheimer’s disease and other dementias, cataract, gallbladder and biliary diseases and atrial fibrillation and flutter (AFF). We considered that costs for AFF may already be included under the other cardiac related conditions costed.

Productivity gains estimates: gains resulting from a reduction in obesity-related morbidity, mortality and obesity-related mortality and morbidity (combined).

We modelled the entire 2019 population in Kenya with risks rising only from age 20 and no burden among children.

*cost offsets are healthcare cost savings. Net monetary benefit was calculated as the total healthcare cost savings and productivity gains realised less total intervention cost.

In Table 6, we calculate the maximum justifiable cost for a research-based strategy that leads to increased production and consumption of healthy indigenous foods. This is the maximum intervention cost that could be incurred whilst ensuring that intervention remains cost-effective or very cost effective.

If healthcare cost savings alone are considered and used to guide on the maximum intervention cost, for the first intervention scenario where there is a change in consumption levels related to supermarket food purchase, the maximum intervention cost that could be incurred whilst ensuring that intervention remains cost-effective is USD 1,924,647,709, for the second intervention scenario on a change in national consumption levels back to the 1975 average levels of intake, maximum justifiable would be USD 6,225,633,667. If considering both the healthcare cost savings and the value of HALYs gained, how much one is willing to pay for a HALY would determine the maximum justifiable cost. Where a threshold of USD 5,403 is applied as value per HALY gained (i.e., 3 times the GDP per capita for Kenya in 2019) [11], for the first intervention scenario, the maximum justifiable cost would be USD 21,795,123,562 while the second intervention scenario would be USD 77,050,705,862. Where a threshold of USD 1,801 is applied as value per HALY gained (i.e., the GDP per capita for Kenya in 2019, for the first intervention scenario, the maximum justifiable cost would be USD 8,548,139,660 while the second intervention scenario would be USD 29,833,991,065.

Table 6 The maximum justifiable cost

Sensitivity analysis

When 5% discounting was applied, the lifetime HALYs gained reduced by ~ 50% for all interventions when compared to base case (3% discount rate). When no discounting was applied the HALYs gained over the lifetime were about 4 times the base case scenario (Table 7). When 5% discount rate was applied, the lifetime healthcare cost savings reduced by about 30% compared to base case. The lifetime healthcare cost savings were 2 times those in base case when 0% discount rate was applied. As expected, compared to base case scenario, the health and economic gains increased when a 30% SSB tax was implemented and reduced when a 10% SSB tax was implemented or when cross price elasticities were applied. Including a 120% pass on rate increased both the lifetime HALYs gained and healthcare cost savings by 16% while an 80% pass on rate reduced the lifetime HALYs gained and healthcare cost savings by 17% (compared to base case) (Table 7). In the sensitivity analysis, the two interventions assessed for cost effectiveness were found to be very cost effective (without cost offsets) and dominant (with cost offsets included).

Table 7 Health and economic impact for various scenarios included in sensitivity analysis

Discussion

Statement of principal findings

Our aim was to assess the impact of the different proposed interventions; taxation on sugar sweetened beverage, mandatory kilojoule menu labelling on food served in formal sector restaurants, change in consumption levels related to supermarket food purchase and change in national consumption levels (return to the 1975 average levels of intake) in control of obesity and overweight in Kenya.

Our findings show that all interventions evaluated would yield substantive reduction in people with overweight and obesity which translated to significant health gains, healthcare cost savings and productivity gains. Over the lifetime of the 2019 population, the impact on HALYs translated to 4, 3, 73 and 261 HALYs per 1,000 persons for a 20% tax on sugar-sweetened beverages, mandatory kilojoule menu labelling, a change in consumption levels related to supermarket food purchase patterns and for a change in national consumption back to the 1975 average levels of energy intake, respectively. There were wide differences in the sizes of the impacts with the two specific interventions (a 20% tax on sugar-sweetened beverages and mandatory kilojoule menu labelling) resulting in good impact while the two based on more broad scenarios that were assumed to result in large changes in consumption patterns, had very large impacts.

For context, from the change in national consumption levels back to the 1975 average levels of intake, the lifetime healthcare cost savings are 1.4 times the 2019 Kenya total healthcare expenditure or 6.2% of the 2019 GDP. Savings from a change in consumption levels related to supermarket food purchase translate to 42% of the 2019 Kenya total healthcare expenditure or 2% of GDP, savings from the 20% tax on sugar sweetened beverages were 3% of total healthcare expenditure or 0.1% of GDP while those from the mandatory kilojoule menu labelling translated to 2% of total healthcare expenditure or 0.08% of GDP. The cost savings realised will not only benefit the government but will also bring relief to many households due to the current high out-of-pocket healthcare costs incurred by individuals in Kenya (~ 28%) [85]. For each intervention, the lifetime productivity gains due to obesity related morbidity and mortality (combined) were about 14 times greater than the corresponding healthcare costs savings.

The intervention costs were 3% and 6% of the projected healthcare cost savings. Translating to about 0.1% of the 2019 total healthcare expenditure in Kenya for both interventions. This is indicative of the feasibility of the two specific interventions in Kenya. The two specific interventions were assessed for cost effectiveness and found to be very cost effective (without cost offsets) and dominant (health promoting and cost-saving) with cost offsets included, i.e., from a health sector perspective.

When various sensitivity scenarios were assessed, both interventions were still found to be very cost effective (without cost offsets) and dominant (with cost offsets included).

Comparison with other studies

To our knowledge, this is the first study to assess the potential economic and health impact and cost effectiveness of interventions for the prevention and control overweight and obesity in adults in Kenya. A similar study has been done by Ananthapavan et al. who assessed the cost effectiveness of obesity prevention policies in Australia [20]. The authors used similar modelling methods. Of the 16 interventions that they assessed, they included mandatory kilojoule labelling on fast food and a 20% sales tax on SSB. They described the kilojoule labelling on fast food as mandatory legislation for fast food outlets displaying energy content of foods and drinks on menus accompanied by a government education campaign. Both interventions yielded health gains and were cost saving. In the assessment for cost effectiveness, the two interventions were found to be dominant (health promoting and cost-saving). The menu kilojoule labelling legislation was found to have high feasibility, sustainability and acceptability by public and government [20]. Other modelling studies carried out in high income countries have also found that a tax on SSBs resulted to changes in average body mass translating to substantial health gains, healthcare cost savings [59, 86], and the tax intervention was found cost effective [87, 88]. Similar studies have evaluated the health and/or economic impact of SSB tax in low-, middle- and upper-middle income countries [89,90,91,92,93,94,95,96]. Some differences in these studies include the varied tax types and thresholds included in the studies, sources of evidence of effect, model decisions/assumptions made, stratification of analyses e.g., by whole population or across income groups and, different health or economic outcomes are assessed. However, across all the studies, a tax on SSBs was found to reduce obesity related morbidity and premature deaths, reduce healthcare costs and to be cost effective.

Strengths and limitations

We apply established modelling methods to assessing the health and economic impact and cost effectiveness of interventions for the prevention and control of overweight and obesity in Kenya [11, 59, 76]. As a strength, our model includes a comprehensive list of high BMI related NCDs as identified in the literature. In the base case analysis, we included extensive uncertainty to account for uncertainty in data and evidence base. We also test some sensitivity scenarios to assess robustness of our study results. In addition, we expanded our study to include the assessment of productivity gains, net monetary benefit and estimation of the revenues that taxation could generate for the government.

Another key strength of our study is that the selection of our interventions was informed by stakeholders from various sectors across government, development agencies, higher education and research and civil society who influence priorities for NCD control in Kenya [26]. Additionally, the model scenarios and evidence of effect for each intervention were underpinned by evidence in literature [40, 54, 55, 58, 60].

Due to limitations of available evidence, the likely impact of kilojoule intake on body weight was based on the assumption that there is no compensatory behaviour (e.g., physical inactivity following reduced consumption). Further research on possibility or extent of compensatory behaviour could enhance the evidence base.

Though all our interventions targeted whole populations, we used available evidence to scale the reduction in BMI associated with two interventions: not purchasing foods from supermarket and with mandatory kilojoule labelling on food served in formal sector restaurants [10, 40]. To estimate the population purchasing foods from supermarkets, we used findings from Demmler et al, who highlighted that their study sample (n = 550) was not representative of the whole country [40]. Nationally representative data on the number of people who make purchases in supermarkets could inform the size of the target population for this intervention in Kenya.

Enforcement of the mandatory kilojoule labelling is more likely in the formal food retail outfits as opposed to the informal sector. Since there is a large informal food market in Kenya, data on estimates of formal verses informal sector could help tailor the intervention further to the Kenya context. Nonetheless, it is important to note that there is a rapid growth of formal sector restaurants, especially fast-food retailers in Kenya particularly in the urban sector where currently most of the overweight and obesity burden is found (25% of those residing in urban areas have overweight compared to 16% in rural areas) [10]. Additionally, urbanisation is spreading rapidly within the rural areas of the country and a rise in BMI is expected there too [56].

The data on disease costs were specific to Kenya and similar settings and estimates for total healthcare expenditure and intervention costs were country specific. However, as detailed in our previous work [67], we encountered some limitations in costing. In summary, our disease cost estimates may be high since identified costing studies were hospital based which reflect cost of treatment for advanced disease cases. We used disease input data from GBD that adopts broader case definitions where prevalent numbers for instance, may include people not aware they have the condition. To limit the overestimation of cost, we used published costs estimates from public facilities as opposed to private facilities. Future costing studies for the eight diseases not costed would enable assessment of the full health and economic benefit arising from implementation of the population-based interventions modelled. On balance, our study is likely to have underestimated healthcare cost savings.

We did not include industry costs in our evaluation due to lack of data on estimates of industry costs that may arise from the two regulatory interventions. This may mean that we underestimated the true cost of the regulatory interventions. However, in most instances, food labels may already exist with other details on them such as price, expiry date. The industry will only be adding information on kilojoules and possibly a health message on the labels using existing technology and this may not be expensive. For the tax on sugar-sweetened beverages, industry is continually reformulating products either voluntarily or due to mandatory measures and often already have funds set aside for reformulation. In addition, advancement in technology has allowed for less costly reformulation processes than seen in previous years. A key limitation was that the intervention scenario on increasing consumption of indigenous healthy foods was excluded from the cost effectiveness analysis. As reflected by the modelled interventions for this (a change in consumption levels related to supermarket food purchase and change in national consumption levels back to the 1975 average levels of energy intake), the indigenous foods intervention would have a gigantic effect. However, there is little or no evidence, and the outlined scenario in Fig. 1 is highly unlikely particularly for the change in national consumption levels. We also were not able to determine the full cost of the hypothetical scenario under the indigenous food intervention. However, we provide a record of the costing process with feasible estimates for as many aspects as was possible.

Implications for policy and future research

Though there are several national health strategies in Kenya that guide the prevention of overweight and obesity and related NCDs [17,18,19], most of the government policy actions are still at the development stage [30]. Our study provides evidence on potential health and economic benefits and cost effectiveness of interventions that Kenya could consider for the prevention of overweight and obesity in the country.

The first intervention is an extensive proposal for research on Kenya’s indigenous foods. We did not identify any previous research that investigates the effect of increased consumption of healthy indigenous foods on either kilojoule intake, body weight or body mass index for a population. This is an area for future research. Nonetheless, our modelled two scenarios indicate that this intervention has potential for substantial health and economic benefits for the entire population.

Implementation of the mandatory kilojoule labelling on food served in formal sector restaurants and tax on sugar-sweetened beverages interventions in Kenya could be complemented by health education campaigns to the public on healthy diets particularly on daily recommendation energy and sugar levels. An aspect to consider would be the combination of the interventions. If the two regulatory interventions were implemented at the same time, the health education could cover awareness creation for both aspects of healthy diet. This also means that potentially more health gains and cost savings would be expected. To model the actual impact estimates of combined interventions, we would require evidence of (combined) effect. Future research that estimates the joint effects could help identify possible intervention packages that policy makers could consider.

Considering the COVID-19 pandemic, recent evidence shows that weight gain was reported during the pandemic [97]. This may lead to a greater proportion of the population having overweight or obesity. The effect of interventions on consumption and BMI would not differ. However, the health impact would be greater because a greater proportion of the BMI distribution is in the ‘danger zone’ in the exponential risk curve.

Though specific considerations and further research on the proposed interventions is required, most evidence will be gathered upon implementation of the interventions and data collection or through experimental or pilot studies in Kenya or elsewhere. Examples of such considerations include determination of whether people read and understand the labels on the foods served in formal sector restaurants, whether different labelling mechanisms are required for dine in customers and those who take away and inclusion of appropriate health messaging including pictorials.

Placing results from cost-effectiveness analyses within a broader framework that incorporates other factors (implementation considerations) that are important to decision-makers improves the relevance to policy and priority setting [20, 28, 77]. Future studies could incorporate the assessment of these factors in relation to our modelled interventions.

Healthy eating has been linked to sustainable environments [98]. Hence, interventions that create healthy food environments in Kenya are also likely to contribute to food security and protect against climate change. Also, additional benefits could be derived if the government used the revenue generated from the SSB tax to improve the health and wellbeing of the citizens. In turn, this could enhance acceptability of the tax intervention by the public and other key stakeholders. Future priority setting studies should include these factors as part of the other implementation considerations.

Subject to data availability, future studies could assess the impact and cost effectiveness of interventions at the county (sub country) level in Kenya and stratify analyses based on socioeconomic factors such as education level, wealth quintiles, urban versus rural residence.

In relation to cost-effectiveness, there is emerging discussion that ICERs should be compared to cost-effectiveness thresholds based on the best estimates of opportunity cost of health care spending and not the consumption value of health [99, 100]. Woods et al. and Ochalek et al. provided indicative estimates of cost-effectiveness thresholds for Kenya on the basis of opportunity costs [99, 100]. However, the estimates were based on limited data and strong uncertain assumptions. The authors rightly attribute this to the lack of attention paid to estimating opportunity cost of health care spending in the literature to date. This is an area of future research.

Stakeholders engaged in our study had proposed a total of 24 broad strategies for the prevention of overweight and obesity in Kenya [26]. Due to time limitation we included the two highest ranked broad strategies in this study. Also, we did not assess any physical activity related interventions in this study. Future research should evaluate additional preventive strategies particularly those that create supportive environments that make the choice of healthier foods and regular physical activity the easiest choice [101]. Such evaluations will provide additional evidence that informs priority setting for NCD control in Kenya and similar settings.

Conclusion

All interventions evaluated in this study yielded health gains, healthcare cost savings and productivity gains. The lifetime productivity gains estimated were greater than the corresponding healthcare costs savings. In both main and sensitivity analysis, the two interventions assessed for cost effectiveness were found to be very cost effective (without cost offsets) and dominant (with cost offsets included), i.e., from a health sector perspective. These interventions should be given consideration for implementation as part of Kenya’s NCD control plans.