Introduction

Inequality describes the uneven distribution of material resources or opportunities within society. Social variables such as gender, ethnicity, education, and income, among others, shape these opportunities and create a gradient of inequality [1,2]. The impact of inequality on health has been widely documented [3] and there is consensus that social, non-medical factors can determine health outcomes, thus referred to as social determinants of health (SDH) [4].

Chronic diseases —where lifestyle plays a strong role in aetiology and recovery— are greatly affected by SDH [5]. Unhealthy lifestyles are often grouped into clusters associated with social structures and living conditions that restrict the availability and accessibility of healthy choices. These social structures are generally marked by differences in gender, ethnicity, income, and social position [6].

There is ample evidence on the impact of SDH on type 2 diabetes (T2DM) prevalence and complications. For example, a systematic review showed that economic inequalities affect the number of complications in people with T2DM [7], and glycaemic control was also found to be affected by SDH in varying degrees [8], while an interaction between socioeconomic deprivation and genetic predisposition might explain the increased prevalence of the condition in people from South Asian and African backgrounds [9]. Therefore, addressing inequalities is emphasised as means to reduce T2DM incidence and improve health outcomes of people with T2DM [10].

T2DM was selected as the focus of this review because it is among the most prevalent chronic conditions globally and one of the most strongly socially patterned, with its aetiology, management, and complications associated to modifiable lifestyle factors that are themselves shaped by SDH [5,6]. T2DM management is largely ambulatory, requiring sustained engagement with primary care, medication adherence, dietary change, and self-monitoring, which makes it sensitive to the access and resource barriers that vary across the social gradient. T2DM is a significant driver of healthcare expense around the world. Direct costs which encompass the expenses associated with the diagnosis, treatment, and management of the condition are projected to cost 1.51 trillion dollars worldwide by 2030. Indirect costs such as productivity losses due to absenteeism or premature death will reach 0.73 trillion dollars globally according to the same estimation [11]. In addition, intangible costs —those related to the burden of living with T2DM— have also been estimated although less frequently as they are much harder to assess. For instance, one study estimated these at $497 (2019 USD) per person per year in a developing country [12].

SDH also determine health service use, although the direction is not universal and is often country specific. For example, in the UK, a steep gradient in inpatient hospital admissions associated with neighbourhood deprivation has been reported. The costs were estimated at £4.8 billion a year which were partially offset by the lower life expectancy of the most deprived groups [13]. While in Sweden, even after adjusting for age and sex, the highest income groups had increased per capita health expenditures [14]. However, in general, the costs of T2DM according to SDH have not been thoroughly assessed.

Given that T2DM imposes a significant financial impact on healthcare systems and this is potentially greatly affected by SDH, this review aimed to analyse published evidence on the economic burden of health disparities as defined by SDH in the treatment of people with T2DM.

While the impact of SDH on T2DM prevalence, complications, and glycaemic control has been systematically reviewed, the economic costs of T2DM according to SDH have not been synthesised. To the best of our knowledge, this is the first systematic review to synthesise the economic costs of T2DM by SDH. Beyond documenting these costs, this review identifies that the direction of cost inequality is moderated by healthcare system structure, generating a conceptual framework that may have broader applicability to chronic disease cost research.

Methods

Results

Included studies

The initial search yielded 5819 unique studies of which 5488 were excluded as they were deemed irrelevant to the research question, 331 studies underwent full review and 22 were selected for analysis. An updated search conducted in October 2025 identified an additional 642 records. Of these, 21 studies were assessed at the full-text stage; however, none met the inclusion criteria, and no additional studies were included in the final analysis. The complete study selection process is illustrated in Fig 1.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. PRISMA flowchart.

https://doi.org/10.1371/journal.pone.0354198.g001

The median sample size was 10575 (range: 1021−32,846). Most studies had a cross-sectional design (68%), and publications dates ranged from 2000 to 2021. Table 1 presents further characteristics of the studies analysed.

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Characteristics of the selected studies.

https://doi.org/10.1371/journal.pone.0354198.t001

Twelve studies evaluated only direct costs [25,26,28,30–35,37,39,41], while seven evaluated both [23,24,27,29,36,38,40]. One study from the latter group also evaluated intangible costs [24]. Some studies further classified direct costs as medical and non-medical costs. The former refers to all medical procedures and physician fees associated with medical treatment while the non-medical category includes cost of transportation, accommodation and meals while traveling to the healthcare centre. Table 2 provides detailed information of what costs and components were analysed in each study and the cost allocation method used.

Download:

• PNG
  larger image
• TIFF
  original image

Table 2. Types of costs, components and costing methods used in included studies.

https://doi.org/10.1371/journal.pone.0354198.t002

Primary outcomes

Findings were conflicting for education, income, and employment, while residence, and ethnicity showed clearer and more sustained trends across studies.

Residence.

Residence status was ascertained by urban or rural settings in 4 studies, all set in the Asian continent [23,27,30,34]. All four studies reported that the direct cost of treating T2DM in rural settings was lower than the treatment of urban patients. Two studies also reported decreased indirect costs [23,27]. The differences ranged from 18% [34] to 71% [27] (Fig 2A). Out-of-pocket (OOP) expenditure was reported in three studies as part of direct medical costs [24,27,30,34]. Guo reported OOP being higher in rural areas [30], Daniel reported higher catastrophic OOP costs in urban areas [27] and Jing reported no difference in OOP costs but the proportion of the OOP relative to their total income was higher for rural patients [34].

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2. Percentage difference according to (A) Residence status, (B) Education, (C) Employment and (D) Income.

https://doi.org/10.1371/journal.pone.0354198.g002

Education.

Data regarding education was mixed. Nine studies provided costs by this SDH. Five provided only direct costs, and four provided both. Additionally, one study from the latter group also provided intangible costs [24]. Regarding direct costs, three studies reported that people with lower educational attainment had higher costs than their higher educated counterparts [24,26,35]. Only crude p values were reported. On the other hand, five studies reported decreased direct costs in people with lower educational level [23,30,32,37,38]. Some studies reported significant differences even after adjusting by other social variables and comorbidities [30,32,37] and pharmacological adherence [37] (Fig 2B).

Intangible costs, defined as the willingness to pay to avoid the physical and psychological strain of T2DM, were 41% lower for less educated people [24] (Table 3, Fig 2B). Four studies reported that lower education was associated with decreased indirect costs, mainly assessed by the loss of productivity of the patient and their carers due to the disease [23,24,36,38]. Three reported significant differences in crude models while one reported 35% less indirect costs but does not provide statistical testing.

Download:

• PNG
  larger image
• TIFF
  original image

Table 3. Summary of percentage cost differences in the treatment of type 2 diabetes by social determinant of health, grouped by region and healthcare model.

https://doi.org/10.1371/journal.pone.0354198.t003

Employment.

Among the studies that reported inequality based on employment status, two found no significant difference in direct costs [30,37]. Direct costs in Bangladeshi patients with T2DM were found to be 5% higher in unemployed patients compared to employed ones (Fig 2C, p < 0.05 in crude analysis) [23]. Interestingly, this study found that housewives and retired patients were those with the highest cost: respectively 21% and 39% higher compared to unemployed patients [23] (p < 0.05 crude between-group comparison). In the US, Hu found significant increased direct costs for the unemployed in a representative sample of adult US patients in both crude and adjusted models [32]. Guo found that retired and self-employed patients had highest direct costs; however, when adjusting the model by other sociodemographic variables, employment was not associated with increased costs [30]. Regarding indirect costs, Endarti et.al. and Afroz et.al. found that unemployed people had significantly lower indirect costs than employed people [23,29] (Fig 2C).

Income.

Eight studies presented cost of T2DM by income (Fig 2D). Four out of the seven studies that reported direct costs found that people with less income incurred lower direct costs [23,30,37,38]. Kim, however, reported that those with less income were associated with significantly higher costs to the universal national health insurance system in Korea [35]. The remaining studies found no significant difference. Regarding indirect costs, there was consistency in reporting lower income associated with significant decreased indirect costs in Bangladesh, Pakistan, and the USA (Fig 2D).

Ethnicity.

Seven studies set in the US analysed the costs of treating T2DM by ethnicity. The median (p25-p75) sample size of the studies was 16,065 (11,841−183,211). White individuals incurred highest direct healthcare expense compared to Black [25,26,28,31,39,41], Hispanic [25,26,28,31,32,39,41] and other ethnicities [26,28,31,39] (Fig 3A).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 3. Percentage difference according to ethnicities (A) White, Hispanic and Black ethnicities, (B) Other ethnicities.

NHB: Non-Hispanic Black, H: Hispanic, NHW: Non-Hispanic White. *p < 0.05 When two values are given the 1^st, one represents crude p value and the 2^nd the adjusted. ns: not significant, nr: not reported.

https://doi.org/10.1371/journal.pone.0354198.g003

Finally, two studies reported costs by ethnic minorities in Canada and in China (Fig 3B). The Canadian study reported that aboriginal First Nation people with T2DM had increased direct healthcare costs compared to non-aboriginal patients [33], with the higher costs driven by 2.74 higher hospital admission rates.

On the other hand, Sun et.al. reported that the Hui ethnic minority in China had significantly lower direct and indirect costs than Han Chinese patients [40]. The indirect costs were explained by the high rates of unemployment and lower income in the Hui while the medical expenditures showed an interesting trend: the authors used quantile regression analysis and found that in the lower quantiles of expense, i.e., healthcare components with lower costs, Hui expenses were significantly lower in crude models but adjusted by education, income and comorbidities, the association was lost. However, in the highest quantile the trend was reversed, and the Hui incurred significantly higher cost than the Han even after adjusting for other variables. Table 3 summarizes these findings and are grouped by region and healthcare model.

Quality of the studies

The quality of the studies plot based is presented in Fig 4.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 4. Traffic light plot showing the risk of bias in each domain analysed according to the Larg and Moss appraisal checklist for cost-of-illness studies.

https://doi.org/10.1371/journal.pone.0354198.g004

Most of the studies were of high quality, with low risk of bias in domains such as the data for cost allocation or the assumptions made for the quantification of resources. The major concerns arose from a lack of detail on the perspective of the study, with only 8 studies reporting this explicitly. Knowing which perspective is taken is essential to ascertain potential biases that might be triggered if the perspective is only that of the cost-bearer. Furthermore, most studies did not state the epidemiological approach, or perform a sensitivity analysis although this might be due to a lack of standardised guidelines for the reporting of cost studies.

Counterfactuals were missing from 77% of studies, mainly because the purpose of the studies were not to assess the cost attributed to T2DM, but the effect of SDH on costs, whether direct or indirect. A counterfactual is a hypothetical scenario where the study population does not have T2DM but is otherwise identical in all other respects [21]. Instead of analysing a counterfactual (non-T2DM scenario), most studies divided their population based on SDH characteristics. This information is collated in Table 2. We did not subtract quality points if a counterfactual was missing.

Discussion

The objective of this review was to assess the cost of inequality in the treatment of T2DM. To achieve this, it is essential to discuss the results in the appropriate context. We note that even though we discuss some results considering the healthcare model of the country in which each study was set, it is not the purpose of this article to discuss healthcare financing as a driver of inequalities.

To interpret the findings of this review, we propose a conceptual framework (Fig 5) comprising two hypotheses that explain how inequalities impact T2DM treatment costs, moderated by each country’s healthcare system structure. These hypotheses are not mutually exclusive and may represent different stages of healthcare engagement among people experiencing inequality.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 5. Conceptual framework illustrating how social determinants of health (SDH) influence the economic costs of treating type 2 diabetes mellitus, moderated by healthcare system type.

https://doi.org/10.1371/journal.pone.0354198.g005

The first hypothesis: Foregone care: people experiencing inequalities incur lower direct medical costs due to their limited ability to afford the time and resources needed for regular medical visits and treatments. This is supported by articles reporting lower direct costs in rural areas in China, India and Bangladesh where costs are usually paid by the patient. Inequality due to rural residence is linked to lower incomes [56] which in turn only permit uptake of lower-cost treatment options [57]. However, the inequality suffered by rural populations is also experienced in healthcare services and infrastructure which might account for the lower direct costs. Specifically, high-complexity hospitals are not available in Chinese rural areas [58], personnel only work during the daytime in Indian rural healthcare centres [27] and, in Bangladesh and India, healthcare infrastructures and workforce have less provision in rural zones [59–61].

Associations with lower income also support this first hypothesis. In particular, countries where costs were found to be lower in the less well-off have mixed models of healthcare with high OOP. Therefore, the lower cost found is most likely to be associated with poor capacity and/or willingness to pay. Health-care utilization depends on the need for care, awareness of the need, willingness, and accessibility of care [62]. Where patients must bear the cost of a disease, the usage of services might be more associated with capacity and willingness to pay than to clinical needs.

Second hypothesis: Accumulated burden. Inequality leads patients to experience higher direct medical costs in the long run, as inadequate prevention and low access to primary care could lead to the earlier onset of severe complications. While overall results were conflicting, less educated patients tended to have higher direct medical costs: 1) in South Korea, a nation with universal healthcare insurance; 2) in the Chinese rural province of Yunnan, one of the poorest areas of Southwest China with 70% of its population working as farmers, where new rural insurance schemes decrease the odds of patients not seeking healthcare due to capacity to pay; 3) in a US study that analysed patients from a managed care organisation. In the last of these, the sample included patients from 8 different healthcare plans but uninsured patients or those who pay out of pocket were not represented [26]. The increased cost observed in illiterate or less educated patients might reflect lower health literacy, as there is evidence of an association between these factors [63]. Higher health literacy is associated with better health outcomes and a differentiated use of health services. Health literate patients use primary care services more [63] whereas less health literate patients use more emergency services [64] which tend to be costlier. The Chinese and US studies found increased costs but did not provide cost component differentiation. This mediation needs to be thoroughly explored in T2DM for a better understanding of the drivers of allocation of resources.

There were, however, other conflicting results. Less educated patients tended to have lower direct medical costs in countries that do not provide universal healthcare access such as India. Higher educated patients might access more and costlier treatment options than those less educated as they are likely to be in higher paying jobs and have easier access to treatments. In South Korea, the cost of the less deprived were also found to be higher. This might reflect genuine clinical need, increased survival, increased severity of the condition or more complications as the lack of access barrier is gone. In countries with a similar healthcare system to the South Korean model, similar results have been found with health service use and costs increasing with neighbourhood deprivation [13,65].

Inequality due to employment also suggests that those facing joblessness incur greater expense, supporting the second hypothesis. In China and Bangladesh there is a gradient where unemployed people incur higher direct costs than those employed. Retired patients have the highest direct healthcare costs in both studies, although this might be due to age and higher odds of costlier complications and comorbidities. Associations with unemployment might be due to the severity of the disease, which was not ascertained in either study, but which could explain higher direct costs. Leaving the labour market due to disease creates a higher risk for patients who are already in vulnerable sectors of the population such as people with low socioeconomic standing [66]. Patients with T2DM have an increased risk of stopping working prematurely and of experiencing unemployment which imposes costs on the economy as a whole [67]. Regarding employment, it is worth mentioning that some specific occupations have been associated with increased risk of T2DM and its complications such as night-shift workers [68], drivers and cleaners [69] but no studies of costs in these populations were found.

The Indian study showing no significant difference in direct costs according to employment status [37] had a source population of predominantly middle- and high-income patients in Delhi; therefore, the unemployed population was not equivalent to those from China or Bangladesh. In Indonesia, direct medical costs were not associated with employment nor income, but lost productivity was related to income. This country is moving towards a universal healthcare system and T2DM treatment costs are covered [29].

Inequality driven by ethnicity was probably the domain with fewest conflicting results. US based studies that assessed ethnicity all agreed that the white people with T2DM had the highest direct and indirect costs. The effect of ethnicity on diabetic healthcare service use and costs in US ethnic minorities has been attributed to a mixture of economic factors and insurance status. However, some studies adjusted for these variables and still found significant differences which might point out to underlying systemic inequalities such as racism in the provision of care to ethnic minorities.

There is evidence that ethnic minorities have increased odds of therapeutic inertia, including delays in initiating insulin when glycaemic targets are not met [70,71]. As insulin is a much more expensive treatment, underlying issues regarding ethnic differences in this variable need to be addressed as might explain decreased healthcare costs in these demographics. People from Hispanic and Black groups have increased odds of worse T2DM outcomes. The extent of the effect of less healthcare resource use and decreased costs in treatment outcomes needs to be further explored.

Two other ethnic minorities were included in the review with different patterns. The Hui Chinese were compared to the majority Han in China. The Hui have shown to have increased prevalence of metabolic syndrome [72] and cardiovascular disease [73]. This might explain why only in the 90^th quintile of costs, the costs are greater than in the Han even after adjusting for confounders. In all other quintiles, the crude cost is lower but is explained by lower education and income which also affects this minority group. Lower cost in lower quintiles might be explained by healthcare foregone owing to delayed healthcare seeking, especially if patients from minority groups must pay for it or have inflexible working hours.

In Canada, the increased costs in the aboriginal minority are potentially explained by the severity of the disease. Besides genetic factors, this population has been generationally isolated and restricted in healthy food choices. Also, their hunting and fishing culture has been impeded which caused a shift in their lifestyle [74]. It is worth noting that, different to the US, in Canada healthcare is free at the point of delivery which might also explain the change of patterns.

These hypotheses may reflect different stages of healthcare utilisation among people experiencing inequalities. The first hypothesis might be more applicable to earlier stages of the condition where preventive treatment is available but underused; while the second hypothesis might take predominance as the condition progresses. Given that studies were mainly cross-sectional, and costs were not ascertained longitudinally it is not feasible to explore this dynamic fully. This highlights the importance of evaluating cost trajectories as well as healthcare resource use alongside costs to get a better understanding where disparities emerge. Therefore, while both hypotheses seem to contradict each other, they might be part of a continuum of healthcare engagement by inequalities that warrants further research.

Regarding indirect costs, it is expected that patients on lower incomes will have lower indirect costs because of worse paid jobs. This is supported by the lower indirect costs found in Bangladeshi rural population as their wages are significantly inferior [75]. Lower indirect costs were also found in unemployed, uninsured and less educated patients. These results are likely driven by fewer missed workdays and the lower cost of each missed workday. Uninsured workers and those with lower educational attainment are likely to be worse paid than those with insurance and will tend to lose fewer workdays because of the higher impact of lost wages. Higher education was associated with higher indirect costs. However, the only study that provided intangible costs, also provided years of life lost due to T2DM whose relationship with educational level is inverse. This highlights that measuring indirect costs, using the human capital approach where income is a main variable or intangible cost measured by the willingness to pay approach, might not provide a thorough picture of the impact of financial costs of T2DM in the life of a patient.

Important policy implications arise from these findings. In countries with high rural OOP burden, expanding rural healthcare infrastructure alone may be insufficient; financial protection mechanisms such as subsidised transport to care or telemedicine reducing travel-related access barriers may be needed to convert foregone care into appropriate utilisation. In universal healthcare systems, the higher costs observed in less educated patients suggest health literacy interventions, particularly around T2DM self-management and primary care engagement, could reduce downstream complication costs. In OOP systems, the same intervention would need to be paired with financial protection, as health literacy alone cannot translate into care-seeking without addressing capacity to pay. As evidence shows a pattern of therapeutic inertia in insulin initiation among ethnic minorities, this becomes a target for quality improvement, monitoring time-to-insulin-initiation by ethnicity within healthcare systems could identify and address a concrete driver of the cost-outcome mismatch identified in this review.

Our review has strengths and limitations. The wide search strategy allowed us to retrieve studies from all over the world which permitted a contextualised analysis of each SDH. We were able to answer to our research question which to the best of our knowledge has not been addressed in a systematic review before. However, we focused our search on peer-reviewed studies which might have left reports from grey literature or from non-indexed articles. No studies from Africa or South America were retrieved, regions with stark inequalities; this might be due to our choices of databases as African and South American studies are less likely to be indexed in the databases we chose [76,77].

The high heterogeneity of studies also warrants some discussion, especially in the components analysed which differ greatly between studies. This was due to the data available and the perspective that informed each analysis. Some studies included dental costs in their calculation of direct medical costs; some included cost of transportation as indirect costs while others considered this a direct non-medical cost. Although out-of-pocket expenditure is the dominant cost-bearing mechanism in several included countries, explicit OOP cost components were not reported. This underreporting is concerning given that OOP expenditure is the metric most directly linked to financial hardship and catastrophic health expenditure at the household level. Standardised guidelines on the reporting of cost studies might help overcome this limitation in the future and improve synthesis of the wealth of information provided. Future cost-of-illness studies in LMIC settings should disaggregate and report OOP expenditure as a distinct cost component.

Not all studies reported statistical analysis by SDH nor adjusted estimates. The lack of adjustment for confounders such as age, comorbidities, or disease severity limits the ability to draw causal conclusions about the independent effect of each SDH on costs. We have indicated throughout Figs 2 and 3, and Table 3, whether each reported difference reflects crude or adjusted analysis, allowing readers to weight findings accordingly.

Overall, this review retrieved more studies from countries with fragmented healthcare systems where healthcare is not universal and not free at the point of delivery. In countries where rural living is highly prevalent, living in rural areas is associated with decreased direct healthcare costs. In countries where healthcare coverage is not universal, being uninsured and from an ethnic minority is associated with decreased direct healthcare costs. As previously stated, this pattern might reflect healthcare foregone, determined by ability and willingness to pay rather than clinical need; however, this needs to be further evaluated. Studies should also assess how different SDHs behave together and impact costs. The gradients generated by unequal wealth and opportunity distribution are rarely unidimensional. People with less access to education are more likely to be unemployed or have low paid jobs which will lead to lower income. This vicious cycle of inequality needs to be further studied from a cost perspective to inform policies. We observe that individuals with higher incomes, better employment, and health insurance tend to incur in greater indirect costs compared to their less well-off counterparts. However, it would be necessary to know how this cost translates into total family income to have a better understanding of the true impact of the disease on the patient.

Future research should prioritise three areas. First, studies from Africa and Latin America are needed to determine whether the foregone care and accumulated burden framework applies in contexts with different healthcare financing structures and SDH gradients. Second, longitudinal studies tracking costs alongside healthcare utilisation over time are needed to test whether foregone care in early disease stages predicts accumulated burden later, which cross-sectional studies cannot capture. Third, standardised cost reporting guidelines for SDH-stratified analyses would improve comparability across studies, particularly regarding the classification of direct, indirect, and intangible costs.

This review makes several novel contributions. It is, to the best of our knowledge, the first systematic synthesis of T2DM costs by SDH. It demonstrates that the direction of cost inequality is not universal but is moderated by healthcare financing model. It proposes a conceptual framework for interpreting cost inequalities in chronic disease that warrants further empirical testing. Finally, it identifies critical geographic gaps in the literature, particularly the absence of evidence from Africa and Latin America, where SDH-driven inequalities are most pronounced.

Inequality is a social and political construct. Societies have many options for organisation, and choosing one approach over others generates divisions of wealth [78]. This review highlights the differences in the economic consequences in the treatment of T2DM by residence, insurance status, and ethnicity. Addressing these disparities is crucial to ensure healthcare delivery is based on true clinical need rather than socio-economic factors.

Supporting information