Assessing a national policy on strengthening chronic care in primary care settings of a low-resource country using patients' perspectives.

Background To improve care for patients with chronic diseases, a recent policy initiative in Thailand focuses on strengthening primary care including training of the team to deliver healthcare based on the concept of Chronic Care Model(CCM). This study conducted a cross-sectional survey of 4,071 patients with hypertension and/or diabetes registered to 25 primary care units and 16 hospital NCD clinics in 11 provinces (76 in total) to examine the effects of provider training and local health systems settings on patients’ perception of the chronic care quality. Methods A home-based interview with questionnaire was conducted on the patients in primary care settings. The questionnaire was adopted from the Thai version of the Patient Assessment of Chronic Illness Care (PACIC+) developed by the MacColl Institute for Healthcare Innovation. The questionnaire contains 20 items from the original PACIC, which measure different parts of the CCM, and an additional 6 items assess the 5A Model. Mixed effect models were employed to compare subscale of patient perception of the care quality between trained upgraded PCUs, upgraded PCUs, ordinary PCUs and NCD clinics. Upgraded PCUs were ordinary PCUs with the multiprofessional team including a physician. Trained upgraded PCUs were upgraded PCUs with the training input. Results Mixed effect models depicted an independent association between every PACIC (as a measure of CCM) and facility type the for of ordinary PCU reporting high to highest scores (ORs: p<0.05) compared NCD This is also the case for patients: the across patients compared patients NCD clinics(p<0.05).


Abstract
Background To improve care for patients with chronic diseases, a recent policy initiative in Thailand focuses on strengthening primary care including training of the team to deliver healthcare based on the concept of Chronic Care Model(CCM). This study conducted a cross-sectional survey of 4,071 patients with hypertension and/or diabetes registered to 25 primary care units and 16 hospital NCD clinics in 11 provinces (76 in total) to examine the effects of provider training and local health systems settings on patients' perception of the chronic care quality.
Methods A home-based interview with questionnaire was conducted on the patients in primary care settings. The questionnaire was adopted from the Thai version of the Patient Assessment of Chronic Illness Care (PACIC+) developed by the MacColl Institute for Healthcare Innovation. The questionnaire contains 20 items from the original PACIC, which measure different parts of the CCM, and an additional 6 items assess the 5A Model.
Mixed effect models were employed to compare subscale of patient perception of the care quality between trained upgraded PCUs, upgraded PCUs, ordinary PCUs and NCD clinics. Upgraded PCUs were ordinary PCUs with the multiprofessional team including a physician. Trained upgraded PCUs were upgraded PCUs with the training input.
Results Mixed effect models depicted an independent association between every PACIC subscale (as a measure of CCM) and facility type with the maximum likelihood for patients of ordinary PCU reporting high to highest scores (ORs: 1.52-1.76; p<0.05) compared to hospital NCD clinics. This is also the case for patients: seeing the same doctor on repeated visits (ORs: 1.66-1.87; p<0.05) or having phone contacts of the providers (ORs:1.42-1.63; p<0.05). Similarly, across all of the 5A model subscales, ORs for patients attending ordinary PCU responded with high to highest scores were 1.4-2.0 times compared to those for patients attending hospital NCD clinics(p<0.05).
Conclusions We could not nd evidence to support effectiveness of the training approach. The training failure might indicate a need to address mismatch between health workforce and workload. It also indicates a need to incorporate delity check into any training program for chronic care addressing the complex healthcare needs.

Background
Low-and middle-income countries (LMICs) do not only face a disproportionately heavy burden of chronic non-communicable diseases (NCD) but also have di culties in scaling-up service delivery models such as the Chronic Care Model (CCM) proven effective in high-income countries. 9,15 So far, empirical data on the application of CCM or other strategies to address healthcare needs of patients with NCD in primary care settings of LMICs primarily con nes to pilot scale or individual studies.
To meet the needs, studies have shown that nancial and infrastructural resources alone are insu cient. LMIC health systems also require human and institutional capacity strengthening to improve the effectiveness, quality, distribution, and continuity of care through smart designs and use of technology. 4 In low-resource settings, adapting disease guidelines requires non-physician clinicians to deliver care and to ensure effective implementation of standardized protocols for diagnosis, treatment and monitoring. , Despite the presence of universal healthcare coverage (UHC) in Thailand over the past two decades, provision of chronic care for patients with hypertension and/or diabetes in primary care settings has faced a challenge of shortage of health workforce especially nurses. 17 To address this limitation, the Ministry of Public Health has taken a policy initiative with a two-pronged strategy: a) allocation of a physician with family medicine training and multidisciplinary health professional as a team to primary care settings at subdistrict of a well-de ned population of approximately 10,000; b) training of the team to deliver healthcare based on the concept of CCM. Our study made use of this policy initiative as an opportunity to address the knowledge gap of up-scaling the service delivery based on CCM in lowresource settings. To do so, we adopted the validated Thai version of the Patient Assessment of Chronic Illness Care (PACIC+) 10 to assess the effects of the policy implementation.

Policy interventions
To improve care for patients with chronic diseases, a recent policy initiative in Thailand focuses on strengthening primary care with the two-pronged strategy. Since 2016, the rst strategy had been applied to an accumulated number of 1137 primary care units (PCU) or 11.6% of the total of 9777 in 74 provinces. [ http://pcc.moph.go.th/pcc/dashboard/?p=teamCount_rpt]. A physician trained in family medicine and new medical equipment (such as ultrasonography, ECG monitor) were distributed to each of the upgraded PCUs. The physician is assigned to provide full-time clinical services of 3-5 days a week to the upgraded PCU in addition to outpatient care services in the referral hospital of the PCU. In contrast, patients seeking care at ordinary PCUs have only one day per week to receive care from the team and a physician with or without training in family medicine.
The training strategy was applied to 21 of the upgraded PCUs in July 2019. From each of the 21 upgraded PCUs, the head and 2-3 clinicians attended two consecutive training workshops (1 and a half days each). The rst one started with a didactic lecture addressing the concepts of the strategy and tools for translating the concepts into practices i.e., system thinking and design thinking. Two small group sessions followed the lecture to discuss experiences and ideas related to the translation of the knowledge tailored to speci c settings. Reading materials focusing on WHO's Integrated People-centered Health Service (IPCHS) and CCM were shared with the participants. The second workshop followed one month after the rst to explore the feasibility and barriers of implementing the strategy making use of the participant experiences. The participants were expected to transfer the knowledge and skills to the rest of the team members in each upgraded PCU. To ensure delity of the implementation theory, follow up support and encouragement throughout the study period were carried out by two implementation support practitioners. They paid a visit to each team of the participants aiming at activating implementationrelevant knowledge, skills, and attitudes, and to operationalize and apply these in the context of the participants. In doing so, they aimed to trigger both relational and behavioral outcomes. For instance, the application of the concept of risk strati cation of the patients was encouraged in order to customized clinical transactions according to the needs of speci c patients instead of treating all patients similarly which usually results in super cial provider-patient dialogue and re lling medications over a period of just 3-5 minutes for each patient. Nevertheless, there was no systematic check of the delity. Patient survey

Population and samples
In order to examine the effects of provider training and local health systems settings (using facility type as proxy), the following number of PCUs and hospital NCD clinics in the same district agreed to participate: 7 trained upgraded PCUs, 6 untrained upgraded PCU, 6 ordinary PCU and 13 hospital NCD clinics ( Fig. 1). The hospital NCD clinics were included since they were supposed to care for complicated patients referred from PCUs. Patients with the chronic conditions attended the participating facilities on their regular visits were asked to take part in the study. In total, 4071 patients gave informed consent and were interviewed at home using the PACIC + questionnaire by trained eld workers during September 2019.
In our study, there were 3 sections in the questionnaire: 1) personal information of the respondents; 2) perceptions of the interactions with providers, and 3) PACIC + items. Supplement provided details of the 3 sections. In brief, the personal information consists of demographic pro le, type of chronic conditions, duration of the conditions and health insurance status. Section 2 explored perception about: channel of contact with providers, receiving care from the same provider on repeated visits.

Statistical analysis
We constructed 2 set of latent variables as subscale from the PACIC + items to re ect the components of chronic care model and 5 A model. The subscales for components of chronic care model (CCM) included patient activation, delivery system, goal setting, problem solving and follow-up and for the 5 A model components included assess, advise, agree, assist, and arrange , . Con rmatory factor analysis was performed using structural equation modeling to evaluate the tness of the data to the PACIC + scale structure. The extent to which the items loaded on to the hypothesized variables and the correlation (Table A1, A2 in supplement) were examined. For CCM subscales, almost all the factors have factor loadings of 0.60 or greater, only 4 items had standardized factor loadings less than 0.6. The goodness of t for the overall model was moderate, and the value of RMSEA and CFI were 0.092 and 0.863, respectively. For 5A the overall goodness of t was slightly lower than that of CCM.
To examine the individual factor and type of primary care setting that associated with patient perception measures, PACIC, each subscale was categorized in to binary variable cut at percentile 75th (0 = low score, and 1 = high score) and treated as the outcome variable. Chi-square's test was performed to explore the association between each independent variable and each outcome and variables that provided p-value of less than 0.10 were included in the multivariable regression analysis. The multilevel regression analysis was considered, as the rst level was individual and the second level was the primary care cluster. Mixed effects logistic regression model was used to examine the association between each subscale and the explanatory variables with a random intercept for "primary care unit (PCU)" level to take into account the correlation among patients in the same PCU. Independent variables that were in the multivariable regression analysis included individual level: sex, age, education (primary, secondary and bachelor), chronic diseases (DM, HT, and both), duration of the chronic disease condition, know about family doctor (yes/no) know health care provider's name (yes/no), type of contact channel (no, mobile phone, Line application, and others), seeing same doctor on repeated visits, (yes/no) and insurance scheme (universal health, social security, civil servant, and others) and type of primary care setting (trained PCU, upgraded PCU, ordinary PCU, and NCD clinic in hospital). Odds ratio and 95% CI were calculated and reported. Stata version 16 (StataCorp. 2019. College Station, TX) was used for the statistical analysis.

Results
Most of the 4071 respondents were: female (73%) aged 59 on average with primary level of education (81%). Almost half (49%) of them were hypertensive and 38% had both hypertension and diabetes with average duration of 7-10 years. Eighty two percent was covered with universal coverage scheme, public health insurance with the largest population coverage.
Using mixed effect modeling adjusted for age, sex, education and duration of the chronic conditions; we found association between CCM subscales and individual patient characteristics or health facility type as follows.
Across the 5 subscales of CCM, ORs for patients attending OPCU responded with high to highest scores were 2-4 times compared to those for patients attending hospital NCD clinics (Table 2). This is also the case for patients: seeing the same doctor on repeated visits (ORs: 1.66-1.87) or having phone contacts of the providers (ORs: 1.42-1.63). Patients with hypertension were less likely to do so as compared to those with diabetes for 2 subscales: goal setting (OR 0.78) or follow-up (OR 0.66). Patients with both diabetes and hypertension were more likely than those with diabetes to make such a report in 2 subscales: patient activation (OR 1.25) and problem solving/contextual counseling (OR 1.26). In contrast, there was no statistically signi cant association between health insurance status and patients' reports of any subscale.  Similarly, the analysis revealed association between the 5A model subscales and individual patient characteristics or health facility type as follows. Across the 5 subscales, ORs for patients attending OPCU responded with high to highest scores were 1.40-2.00 times compared to those for patients attending hospital NCD clinics (Table 3). This is also the case for patients: seeing the same doctor on repeated visits (OR: almost 2) or having mobile phone contacts of the providers (ORs: 1.39-1.62) as compared to those without any contacts. In the opposite, patients with other contact channels were less likely to do so (ORs: 0.55-0.62). There was no statistically signi cant association between health insurance status and patients' reports of any subscale.

Discussion
In contrast to other studies in developed country settings, our ndings did not support the training bene ts of primary care providers on caring for patients with the chronic conditions using CCM. Thom  Apart from the seemingly limitations in the training approach identi ed by our study, imbalance of health workforce against workload could be a major barrier to scaling-up quality improvement of chronic care.
Using a multi-professional projection approach for Thailand, Pagaiya N et al highlighted a severe shortage of nurses in 2026 whereas the supply of doctors, pharmacists, and physiotherapists is likely to be surplus. In primary care settings, the study identi ed the proportion of workload as 100% for nurses and 20% for doctors or other health professionals. Hence, the shortage of nurses might explain di culty to improve quality of care for chronic diseases especially for those upgraded PCUs with higher number of registered population (10,000 per PCU) than that of ordinary PCUs(< 10,000 per PCU). This notion is supported by bigger ORs for the PACIC scores or 5A-model scores reported by the patients seeking care at ordinary PCUs than those at other facilities (Tables 2 & 3).
Self-management support in chronic care could be enhanced by mobile ICT tools such as telephone or online applications as indicated by accumulated evidence from RCTs or systematic reviews. Our study provided evidence indicating the bene t of mobile telephone to support self-management of chronic care in a large scale (Tables 2 & 3). Compared to patients without any mobile contact channels, those with mobile phone contacts were more likely to give a high to highest PACIC scores or 5A model scores, ORs ranging from 1.4-1.7, p < 0.05. In developed countries reported gures of citizens lacking basic digital skills in terms of digital literacy were around 40% in Europe and the U.S. Based on our ndings, it is implicated that mobile phone should be the rst choice for the application of mobile ICT to support selfmanagement in chronic care.
With ORs close to 2 for the patients seeing the same doctors on repeated visits as compared to those without (p < 0.05) based on PACIC scores or 5A model scores, our study supported the importance of continuity of care viewing from patients' perspective.
Finally, with regard to equitable access to quality chronic care, our study ndings of no association between health insurance status and the scores on PACIC or 5A model scale (ORs: 0.70-1.93; p > 0.05) render support of the effect of universal healthcare coverage in lling the inequity gaps. Nonetheless, concern about inadequacy in the power of tests could not be excluded.

Conclusions Limitations
With female patients over represented in our study, the results could hardly be applied to male patients. Causal inference is problematic given the cross-sectional design of our study. We did not account for conventional patient outcomes such as blood pressure, HbA1c, adherence to medications. Despite the limitations, the sampling design involving primary care facilities in vast areas and the large number of respondents enable us to assess the possibility of scaling-up policy interventions on quality improvement of chronic care using the validated standardized tools (the PACIC+) in a middle-income country with UHC. Given the paucity of evidence like this in low-resource settings, our study has made an important contribution to ll the knowledge gaps in scaling-up evidence-based approaches to strengthen chronic care model in primary care setting of middle-income countries with UHC.

Policy and practice implications
Policymakers might nd the training approach insu cient for strengthening chronic care at primary care setting with overburdened service load. The failure might also indicate a need to incorporate delity check into any training program dealing with chronic care aimed at addressing the complex healthcare needs. In addition, PACIC + might be useful to assess and monitor the progress in nationwide implementation of the chronic care development in primary care settings. Further studies with more rigorous designs such as effectiveness trials or real world implementation trials are needed to ascertain the effectiveness of training or other approaches using both patient assessment and conventional patient outcomes as indicators. Availability of data and materials

Abbreviations
The datasets generated and/or analysed during the current study are not publicly available as they contain sensitive information and data could potentially identify participants.
Ethics approval and consent to participate  Flow of activities on sample selection and data collection

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download. AdditionalFile1.docx