Introduction

The prevalence of chronic conditions is rising globally, and many countries in sub-Saharan Africa are now facing the double burden of infectious and non-communicable diseases (NCDs) [1]. In Kenya, NCDs account for almost a third of all deaths and this proportion is projected to rise by over 50% during the next decade [2]. Cardiovascular diseases and cancers are among the most common causes of morbidity, after infectious diseases, and in 2015, about a quarter of the population had hypertension and 5% had diabetes or impaired fasting glycaemia [2]. Availability of screening, early detection and management of NCDs in primary care settings in Kenya is limited. Although strengthening primary health care is a priority [3], the health system has remained hospital-centric, with long waiting times and reduced quality of care [4], resulting in poor retention in care of people who screen positively for NCDs and low treatment adherence [5,6].

Availability of and access to services are important for the successful management of NCDs. However the social and economic contexts within which people with NCDs live also greatly impact their capacity to seek care and self-manage their conditions [7]. Evidence from Malawi and South Africa has pointed to the struggles people with diabetes and/or hypertension face on a daily basis [8–10]. For example, Matima et al. explored the experiences of people with HIV and diabetes comorbidity using the cumulative complexity model developed by Shippee and colleagues, which emphasises how clinical and social factors accumulate and interact to complicate patient care [9,11]. They identified two sets of workloads people have to deal with: ‘clinic-related’ workload around lack of service integration and perceived power imbalances between the patient and the health care provider, and ‘self-care’ related workloads around nutritional needs, medication burden and stigma. Available evidence suggests that a high (perceived) burden resulting from patient workloads may negatively affect adherence to treatment and quality of life, in particular among people with multiple and chronic conditions [12–15]. These challenges are further exacerbated by financial concerns, with households in Kenya facing a significant economic burden associated with NCD diagnosis and treatment costs [16].

Much of the work on patient work and treatment burden has been conducted in high-income countries [17,18] and similar work is only beginning to be undertaken in low- and middle income countries [19]. This study seeks to contribute to this emergent evidence by exploring the treatment and self-management burden among people with chronic conditions, in particular hypertension or diabetes, in Kenya.

Methods

This study was set in the context of a wider implementation study that sought to understand the impact and scalability of a novel approach to integrate promotive, preventive, and curative care for diabetes, hypertension, cervical and breast cancer at the primary health care level (Primary Health Integrated Care Project for Chronic Conditions, PIC4C) within the Academic Model Providing Access to Healthcare (AMPATH) programme in Western Kenya [20]. PIC4C was launched in 2018 by the Kenyan Ministry of Health in partnership with AMPATH/Moi, Access Accelerated and the World Bank. The model was piloted in Busia and Trans Nzoia counties in Western Kenya, which formed the location for the present study.

One key component of our study was to understand the treatment and self-management burden among a group of people diagnosed with hypertension and/or diabetes since PIC4C implementation and how the burden is distributed across a range of socio-demographic characteristics. We used the Patient Experience with Treatment and Self-Management (PETS) instrument [21] as it assess treatment burden in patients with chronic health conditions requiring self-management. This comprehensive measure allows understanding what aspects of treatment and self-management prove to be most burdensome to people with chronic conditions in Kenya. The survey served as a basis from which to recruit a subsample of people with hypertension and/or diabetes to further explore treatment burden using in-depth interviews.

The PETS builds on a conceptual measurement framework developed from interviews and discussions with people with multiple chronic conditions in the USA [22] and has since been used in a range of populations with chronic conditions in the USA and Norway, including multiple conditions [14,21], cancer [23,24], heart failure [25], diabetes [26], and hypertension [27]. To our knowledge, this patient-reported measure of treatment burden has not yet been applied in a lower middle-income country.

Results

Table 1 presents socio-demographic and health-related characteristics of our patient sample by county. Almost two-thirds of respondents (63%) were female; the mean age was 57 years (range 20–90) and 73% were married. The two county samples differed significantly on several variables, including ethnic group, educational attainment, and current working. For example, while a large proportion identified as Luhya in both counties (57% in Busia and 48% in Trans Nzoia), the second largest group in the Busia sample identified as Teso (33%) whereas in Trans Nzoia sample, about 22% identified as Kalenjin and 19% as Kikuyu. The Busia sample also had a higher proportion of respondents with no or only primary education (78% vs. 60%), and a smaller proportion that reported to be currently working (78% vs. 90%). Among those currently working, over half of the Trans Nzoia sample reported to be self-employed in agriculture (55%) compared to 39% in Busia. Average household income was low in both counties, with about half of respondents reporting a monthly income of less than KShs 3,000 (US$27; £20; €23 in January 2021).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Socio-demographic and health-related characteristics of the patient sample by county.

https://doi.org/10.1371/journal.pgph.0001407.t001

The majority of respondents reported having one chronic condition, with the Busia sample having a significantly higher proportion of respondents with two or more conditions (48% vs. 30%). Over 80% in each county reported having hypertension and between 37% (Trans Nzoia) and 44% (Busia) had diabetes; the mean number of reported chronic conditions was 1.6 for the Busia sample and 1.3 for the Trans Nzoia sample. Only about a quarter reported having had NHIF cover at the time of the survey in either county.

Table 2 shows reliability of PETS domain scales for the sample overall and by county (S1–S5 Tables show the frequency of responses to individual PETS domain items). Internal consistency reliability was generally good for all multi-item scales, and Cronbach’s coefficients were well above the threshold for adequate reliability (α ≥ 0.70). The only exceptions were ‘monitoring health’ (α = 0.55), ‘medical and health care expenses’ (α = 0.69) and ‘diet’ (α = 0.67), although the latter two were close to the adequate reliability threshold. Monitoring health asks about the ease or difficulty of monitoring health behaviours (e.g. tracking exercise, foods eaten, or medicines taken) and health condition (e.g. weighing, checking blood pressure or blood sugar levels). Our cognitive interviews found that while respondents generally found the question itself easy to understand, some required further explanations of what was meant by ‘monitoring’. Furthermore, the ability to track, say, blood pressure or blood sugar levels at home requires respondents to have the relevant equipment at their disposal but, as noted earlier, only 5% of survey respondents reported using medical equipment that would enable self-monitoring at home.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Reliability (Cronbach’s α) of PETS domains scales.

https://doi.org/10.1371/journal.pgph.0001407.t002

Lower internal consistency reliability for the domains medical and health care expenses and diet was largely driven by the Trans Nzoia sample (Table 2). The former domain includes one question about the ease or difficulty to understand what is and what is not covered by health insurance. Leaving this item out would increase Cronbach’s α for the domain to 0.918 in the Trans Nzoia sample (0.889 in the Busia sample). The item had a relatively large number of ‘not applicable’ responses (20% in Busia, 25% in Trans Nzoia). As only 25% of sample reported having NHIF cover it is conceivable that the item was not relevant to most respondents, although we have retained it in our subsequent analysis as removing it did not change findings in any discernible way. The domain diet includes the item ‘It is hard to find healthy foods’, which returned a negative Cronbach’s α when deleted for the Trans Nzoia sample, possibly reflecting the limited spread of answers in this sample across three categories (‘strongly agree’, ‘agree’, ‘disagree’) only. Finally, the domain difficulty with health care services showed good internal consistency reliability for the total sample but not for the Busia sample (α = 0.56). This might be explained by the relatively large proportion of ‘not applicable’ responses for three of the items in this domain (S1 Table).

Table 3 presents descriptive statistics of 11 PETS domains and two single item scores for the total sample and by county. The highest treatment burdens were reported in the domains of medical and health care expenses, monitoring health, and physical and mental exhaustion related to self-management as well as diet and exercise/physical therapy. Several items within other domains were also rated as especially burdensome, such as finding transport to get to (S1 Table) and long waits at medical appointments (S2 Table), feeling dependent on others for health care needs (S3 Table) or self-management interfering with work, family responsibilities or daily activities (S4 Table). Fig 1 disaggregates domains with the highest mean burden scores by response item and county.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Response frequency for selected PETS domain items, by county.

https://doi.org/10.1371/journal.pgph.0001407.g001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Descriptive statistics of 11 PETS domain scale scores and two single-item indicators.

https://doi.org/10.1371/journal.pgph.0001407.t003

Compared with the Busia sample, Trans Nzoia mean burden scores were significantly higher for medical information, medications, medical appointments and difficulty with health services. Conversely, Busia respondents reported a significantly higher burden for interpersonal challenges, medical expenses, role/social activity limitations and diet (Table 3). Scale scores were positively skewed toward a lower burden in most domains except for medical expenses, which was slightly negatively skewed toward a higher burden. Diet was also negatively skewed in the Busia sample. Floor effects were generally lower in the Trans Nzoia sample.

There were significant differences in mean burden scores by ethnic group in the domains medications, medical appointments and difficulties with health care services, with those identifying as Teso reporting a substantially lower burden. For example, in the domain difficulties with health care services, Teso respondents had a mean score of 21.3 (SD 20.0) compared with a score of around 40 for the other three main ethnic groups (S6 Table). At the same time, Teso respondents reported a significantly higher burden in the relationship with others and medical expenses domains (S6 Table).

There were generally fewer differences in mean burden scores by age, gender, education, income or number of chronic conditions. Younger respondents (<60 years) reported a significantly higher burden for bother with medication side effects and for diet (p < .05); they also reported a higher burden for medical appointments, which was borderline significant (p = 0.075). Respondents with no formal education or those on low income also reported a significantly higher burden for diet as well as for medical expenses (S6 Table). Perhaps not surprisingly, those reporting having health insurance cover reported a significantly lower (albeit still comparatively high) burden for medical expenses compared to those without health insurance cover (53.1 (SD 25.7) vs. 66.3 (SD 21.1).

Fig 2 disaggregates PETS scores for the total sample by number of chronic conditions. Perhaps somewhat counterintuitively, we found higher scores for people reporting one chronic condition compared to those with two or more conditions in several domains although differences were not statistically significant. The only domains where burden scores were significantly higher with more chronic conditions were role/social activity limitations and diet. There was a clearer relationship between the number of drugs taken and reported treatment burden in most domains although differences were significant for bother with medicine reliance, bother with medication side effects, role/social activity limitations and diet only (S6 Table).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. PETS scores by number of chronic conditions, total sample.

Note. * role/social activity limitations: p = 0.005; diet: p = 0.030. See also S7 Table.

https://doi.org/10.1371/journal.pgph.0001407.g002

Exploring treatment and self-management burden for type of chronic condition, we found that people with diabetes (with or without other conditions) reported a higher burden in all domains compared to people with any other NCD (including hypertension), and a significantly higher burden for medical appointments, role/social activity limitations, bother with medication side effects, and diet (S6 Table).

Discussion

To our knowledge, this is one of the first studies that have used a patient-reported measure of treatment and self-management burden among people with chronic conditions in a lower middle-income country. We found that patients in Western Kenya who were recently diagnosed with hypertension or diabetes reported a considerable treatment burden in a range of areas, with particularly high burdens around difficulty paying for health care, monitoring health, and physical or mental exhaustion from self-management, alongside affording or following a healthy diet and engaging with exercise or physical therapy. Other areas perceived as especially difficult or bothersome included finding transport to get to and long waits at medical appointments, feeling dependent on others for health care needs and the impact that self-management had on work, family responsibilities or daily activities.

The reported treatment and self-management burden differed between patient populations in the two counties in most domains, and this appeared to be driven by a combination of factors, including ethnic group, educational level and burden of multiple chronic diseases. However, the group sizes were too small to allow for further robust analysis of underlying patterns.

Empirical application of the PETS has so far only been documented for populations in the USA and Norway, and the reported treatment and self-management burden is consistent with our findings in so far as the highest burden among people with diabetes or multimorbidity was in the domains of medical expenses, monitoring health and physical or mental exhaustion from self-management [21,33]. PETS scores in US patient populations tended to be lower than in the Kenyan samples queried in this study, although in Eton et al.’s 2017 study, a subgroup of participants recruited from an urban safety-net hospital (Hennepin County Medical Center in Minneapolis, Minnesota), which provides care for low-income, uninsured, and vulnerable persons, had mean PETS scores that were much closer to those reported in our sample [21]. An additional study by Eton et al. [34] developed and administered a briefer short-form version of the PETS with the same patient population and found even higher mean scores in several burden domains. However, caution must be exercised when making direct comparisons given the different version of the PETS used in this prior study. Contrary to other measures of treatment burden, such as the Treatment Burden Questionnaire (TBQ) [35] or the Multimorbidity Treatment Burden Questionnaire (MTBQ) [13], psychometric testing of the PETS has not supported a global summary burden score, although more recent work has distinguished ‘workload’ and ‘impact’ summary scores which aggregate some of the PETS domain scales [14]. Determination of severity thresholds for PETS domain scores (i.e., low, medium, and high burden) are currently pending and may make comparisons with scores from other measures more feasible in the future.

Overall, our findings for treatment burden in a population of people with diabetes and/or hypertension in Western Kenya generally align well with other studies that used comparable measures. Work that has assessed treatment burden in different populations found that younger people [13,34,36], those with greater financial difficulties [13,21,35], and those with multiple conditions tended to report higher treatment burdens [14,21], although the international evidence is somewhat mixed on the latter [13,14,35]. We also showed treatment burden to be significantly associated with younger age although this was in selected domains around medical appointments, bother with reliance on medicines and diet only. Duncan et al. argued that a higher burden might reflect role differences, with younger people having to organise medical appointments around work commitments and looking after dependents while perhaps also having different expectations in terms of managing their own health [13]. Similar to studies in the USA, lower income was associated with a higher treatment burden in our sample, in particular around medical expenses [14,21]. Studies of populations in countries with universal health systems did not find such associations [13,35,36]. Moreover, we found that respondents who had health insurance cover reported a significantly lower (albeit still high) treatment burden as it related to medical expenses. Taken together, these observations highlight the importance of financial safety netting not only to protect people from financial risk related to managing chronic conditions but also to lower the associated treatment burden.

We did not find treatment burden to be higher among people with multiple chronic conditions except in the domains bother with medicine reliance, role/social activity limitations and diet. Possible explanations for an apparent lack of association between treatment burden and number of chronic conditions include that those with more than one condition might find it easier to call upon and navigate medical and social support because they are more experienced and ‘already in the system’ [37], and, possibly, because of the integrated provision of services within PIC4C, which would otherwise have required repeat visits to different clinics. We were unable, in this study, to assess disease severity; we did, however, find a dose-response relationship between the number of drugs taken and treatment burden, which could be indicative of greater perceived or experienced severity. Perhaps not surprisingly, we found that people with diabetes reported a higher treatment burden compared to those with hypertension. While it is difficult to compare directly, studies of people with (multiple) chronic conditions in Switzerland (using the TBQ) [38] and Victoria, Australia (MTBQ) [39] also found a positive association of treatment burden with diabetes. Herzig et al. [38] suggested that a perceived high treatment burden for diabetes might reflect the wider range of activities that patients have to engage with to effectively manage the condition, from regular drug intake to adapting diet and exercise, all impacting on perceived quality of life. Evidence from low resource settings specifically points to the key challenges of affording and accessing a healthy diet among people with diabetes [40,41].

Supporting information

Acknowledgments

The authors wish to express their gratitude to all patients who have so generously given their time to complete the survey and to Alex Njumwah, Elizabeth Khisa, Kenneth Rotich, Kevinah Asigi, Ruth Nehema, Shameem Mutalib and Topista Nafula, who have conducted the fieldwork.