Ethnic inequalities in patient satisfaction with primary health care in the UK: evidence from recent General Practitioner Patient Surveys (GPPS)

Aims and objectives This paper aims to improve understanding of factors that contribute to persistent ethnic disparities in patient satisfaction in the UK. The specific objectives are to (i) examine ethnic differences in patient satisfaction with their primary care in the UK; (ii) establish recent trends in patient satisfaction by ethnicity; and (iii) examine factors that contribute to ethnic differences in patient satisfaction. Data and methods The study is based on secondary analysis of recent General Practitioner Patient Survey (GPPS) datasets of 2019, 2020 and 2021. Descriptive bivariate analysis was used to examine ethnic differences in patient satisfaction across the three years. This was followed with multilevel linear regression, with General Practice (GP) at level-1 and Clinical Commissioning Group (CCG) at level-2 to identify factors contributing to ethnic differences in patient satisfaction. Results The findings show consistent negative correlations between the proportion of patients reporting good (very or fairly good) overall experience and each of the ethnic minority groups. Further examination of the distribution of patient satisfaction by ethnicity, based on combined ethnic minority groups, depicted a clear negative association between ethnic minority group and patient satisfaction at both GP and CCG level. Multilevel regression analysis identified several service-related factors (especially ease of using GP website and being treated with care and concern) that largely explained the ethnic differences in patient satisfaction. Of all factors relating to patient characteristics considered in the analysis, none was significant after controlling for GP service-related factors. Conclusions Ethnic minority patients in the UK continue to consistently report lower satisfaction with their primary health care in recent years. This is largely attributable to supply (service related) rather than demand (patient characteristics) factors. These findings have important implications for health care system policy and practice at both GP and CCG levels in the UK.

6 147 Furthermore, an examination of barriers to presenting cancer symptoms to GPs among women in 148 the UK identified emotional barriers as prominent among ethnic minority groups [20]. Women from 149 minority ethnic groups were more likely to pray or prefer their traditional forms of medicine. If an 150 individual determines prayer and traditional medicine to be better remedies for cancer symptoms in 151 comparison to the expertise of GPs, then one can assume that the individual may not highly rate 152 their experience of GP consultations since they may not fully appreciate the medical science. A 153 study in South Africa by Munyewende

181
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The Data and variables
212 This paper is based on secondary data analysis of the General Practitioner Patient Survey (GPPS) 213 datasets. The GPPS is nationally-representative repeated cross-sectional survey. It is one of the 214 largest surveys of millions of registered NHS patients invited to disclose their experience in relation . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) where Y ij is patient satisfaction (percent with good overall satisfaction) for GP i in CCG j; 274 B 0 is the regression constant/intercept (average patient satisfaction when all independent 275 variables are zero); X 1ij -X kij are ethnicity and other independent variables considered in the . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 21, 2022.
283 For the multilevel analysis, careful preliminary analysis was undertaken to determine the key

302
Ethical considerations 303 We wish to declare that research reported in this article meets the standard ethical requirements.
304 The study is based on secondary analysis of the GP Patient Survey which has been designed to give 305 patients the opportunity to give feedback about their experiences of their GP practice. It is carried 306 out by Ipsos MORI, on behalf of NHS England. Ipsos MORI is a registered and independent survey 307 organization that strictly adheres to the Market Research Society's ethical code of conduct (http:// 308 www.gp-patient.co.uk/faq). Furthermore, the dataset analysed in this paper was already aggregated . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 21, 2022. 309 at GP level, so it is impossible to link data to a particular individual. Ethics approval was not 310 required for this study since we used anonymised data aggregated at the GP level and routinely 311 available in the public domain.   . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 21, 2022. ; https://doi.org/10.1101/2022.06.20.22276629 doi: medRxiv preprint 361 362 Analysis of the distribution of the main outcome variable on patient satisfaction (percent reporting 363 very good or fairly good overall experience) shows that overall satisfaction has remained 364 consistently high during the last three years (Table 3). Although the satisfaction levels seem fairly 365 consistent across the three years, with 82-83% of the respondents reportion overall good experience, 366 the non-overlap in 95% confidence intervals suggest significant differences (Field 2009 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.  387 388 Overall patient satisfaction by ethnicity is more clearly presented in the stacked bar charts in Figure   389 1, obtained after classifying both ethnic minority and patient satisfaction proportions by quartiles. 418 Further preliminary analysis involved use of scatter plots to better understand the nature of the 419 relationship between patient satisfaction and ethnicity, at both GP and CCG levels (Figures 2 & 3).  Table 5. 445 Table 5  . This suggests that the observed lower satisfaction among minority 467 ethnic groups was partly due to difficulties in using their GP practice's website to look for 468 information or access services. Although always seeing preferred GP was a significant predictor of 469 patient satisfaction, introducing this variable in the model resulted in minimal reduction in the effect 470 of ethnicity, suggesting that it was not a particularly important factor in explaining the lower 471 satisfaction among ethnic minority patients. Being treated with care and concern was a strong 472 predictor of patient satisfaction, and explained a substantial proportion of ethnicity effect (See 473 Annex ii). As expected, being treated with care and concern was associated with significantly 474 higher satisfaction, and was an important factor in explaining the lower satisfaction among ethnic 475 minority groups. Confidence of managing condition was a significant factor in patient satisfaction 476 and explained most of the effect of long-term health condition, and some ethnicity effect. Finally, 477 both trust in healthcare professional and involvement in decisions about treatment were associated 478 with higher patient satisfaction, and together with the factors outlined above, fully accounted for the 479 ethnicity effect in patient satisfaction. These GP service factors also largely explained the observed 480 CCG level variations, as shown by the reduction in ICC in Model 3. Once these were controlled for, . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 21, 2022. ; https://doi.org/10.1101/2022.06.20.22276629 doi: medRxiv preprint 481 only about 2% of the total unexplained variation in patient satisfaction was attributable to 482 unobserved CCG level factors.
483 In addition to the factors presented in Table 5 above, the multilevel regression analysis considered a 484 number of factors relating to patient characteristics including age and working status, identified in 485 the literature to moderate/mediate the relationship between ethnicity and patient satisfaction. There 486 was no evidence that the main effect of these factors (e.g percent aged 65+, percent in paid FT/PT 487 work, etc), nor their interactions with ethnicity were significant. Therefore, these factors were not 488 included in the models presented. 497 However, the reasons for the lower patient satisfaction score among ethnic minorities is yet to be 498 fully determined and is where the main contribution of this study lies.
499 In our predictive analysis based on multilevel linear regression modelling, we considered a range of 500 factors, including patient characteristics and service-related factors that were presumed to be 501 potentially important based on previous research/literature. We identified one demand/patient 502 related factor (long term health condition) and a number of supply/service related factors (ease of 503 using website; always being able to see preferred GP; being treated with care and concern; 504 confidence of managing health condition; trust in Health care professional; and being involved in 505 decisions about treatment) as significant factors that account for observed ethnic differences in 506 patient satisfaction at GP level. 508 Long term health status was the only significant patient-related factor in our analysis. Previous 509 studies suggest that patients' long term health status is likely to affect their satisfaction with care.
510 For instance, Detollenaere et al. [7] noted that patients with long term health issues are likely to 511 score either extremely positively or negatively in patience satisfaction as they are likely to be either 512 appreciative or frustrated with their constant interactions with the primary care providers. Since 513 ethnic minorities are more vulnerable to long-term health issues [30], it was important to establish . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 21, 2022. ; https://doi.org/10.1101/2022.06.20.22276629 doi: medRxiv preprint 514 the extent to which long-term health issues may have been a factor in ethnic disparities in patient 515 satisfaction. When we introduced the variable for long term health condition into the model, it had 516 an effect on the impact of ethnicity in the overall model. Although we had expected the impact of 517 ethnicity to reduce with the introduction of long-term health status, it increased instead.
518 Investigating further, we ran a correlation analysis of long-term health status and ethnic minority 519 variables. The result was a significantly high negative correlation, suggesting that the ethnic 520 minorities who participated in the study were less likely to report having a long-term health 521 condition. When we run correlation analysis between the different age groups and long-term health 522 status, the older age groups had a significant positive correlation with long term health status. This 523 was expected as older patients are most likely to report long term health conditions. A correlation 524 analysis between Ethnic minority and the different age groups indicated that ethnic minorities who 525 participated in the study were relatively younger than their White counterparts. Therefore, it was 526 apparent that older patients were more likely to have a long health term condition, while ethnic 527 minorities were predominantly made up of younger age groups. Long term health status did not 528 account for ethnic differences in overall satisfaction but when factored in it did affect the 529 relationship between ethnicity and overall satisfaction. This did not support our hypothesis that 530 Ethnic minorities would be susceptible to long term health issues, and more likely rate their overall 531 satisfaction more extremely, either positively or negatively, depending on whether they are 532 appreciative or frustrated with their health care. 533 534 We considered introducing several other variables in the model, but they were not significant and 535 therefore excluded. Among these variables were variable pertaining to age and working status. In 536 previous studies age was determined to be a factor in ethnic differences of patient satisfaction but 537 there was no evidence to support this in our analysis. As noted in the literature review, the 538 relationship between ethnicity and patient satisfaction can be moderated or mediated by a range of 539 demographic factors, leading to rather complex patterns. Interactions between various socio-540 economic (e.g percent in paid work) and demographic factors (e.g percent aged 65+) with ethnicity 541 were considered in the regression model but there was no evidence that any of these were 542 significant.

Supply/ service related factors
544 The (Ease of using Website) variable was included as a covariate in our multilevel regression 545 analysis as it was one of the few viable variables that could be used as a proxy for access to health 546 services. As noted by Kontopantelis et al.[11], good access is critical to good health care. Our 547 hypothesis was that ethnic minorities would lack good access to health care most likely due to lack . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 21, 2022. ; https://doi.org/10.1101/2022.06.20.22276629 doi: medRxiv preprint 548 of financial resources [31] and this may affect their overall patient satisfaction. Individuals with 549 access to devices such as laptops, high spec smart phones, tablets etc will be able to receive the 550 maximum benefits of the online services available which is likely to result in a better overall patient 551 experience and satisfaction. A larger percentage of ethnic minorities fall within the lower socio-552 economic group and may not be able to afford such devices. As expected, when we introduced the 553 (Ease of using Website) variable into our multilevel regression model, the effect of ethnicity on 554 overall satisfaction decreased substantially, suggesting that service accessibility was an important 555 factor in partly explaining the lower satisfaction among ethnic minority patients. 556 We had stipulated that ethnic minorities, especially first-generation immigrants, would feel more 557 comfortable seeing a GP who can speak the same language and share the same cultural norm, 560 When the (Always seen preferred GP) variable was added to the multilevel regression model, the 561 effect of ethnicity on overall satisfaction further decreased as expected, although the decrease was 562 marginal. We had further hypothesized that ethnic minorities would be more likely to negatively 563 rate the 'Care and concern' they received predominantly due to cultural differences and 564 communication barriers that would lead to misunderstandings and frustrations between patients and 565 the GP staff. When the (Care and concern) variable was added to the model, the effect of ethnicity 566 on overall satisfaction further decreased as expected. The decrease was substantial, suggesting that 567 this was an important factor in explaining the overall low satisfaction among ethnic minorities.
568 An earlier study conducted by Williams et al. [20] to determine ethnic differences in barriers to 569 presentation of cancer symptoms in primary care among women in England concluded that ethnic 570 minority groups were more likely to pray about a symptom or rely on traditional remedies. Some 571 ethnic minorities, especially first-generation immigrants, are more likely to rely on traditional forms 572 of medicine or prayer instead of the primary care offered as a solution to health issues. They would 573 be more confident in self managing their long-term conditions and may not appreciate the primary 574 care offered nor have trust in health care professionals. Therefore, we had hypothesized that ethnic 575 minorities would be more confident in self managing their condition (through prayer and traditional 576 medicine) and have little trust in health care professionals that would result in lower overall 577 satisfaction. When the variables (Confidence in managing condition) and (Trust In health care 578 Professional) were introduced in the model, there was a decrease in the effect of ethnicity in overall 579 satisfaction as expected, but the decrease was minimal.
580 Schinkel et al.[14] conducted a study in the Netherlands to determine the perception among Turkish 581 -Dutch migrants in relation to barriers in their involvement in primary health care consultations.
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted June 21, 2022. ; https://doi.org/10.1101/2022.06.20.22276629 doi: medRxiv preprint 582 The study noted that the Turkish -Dutch expected a larger power distance in medical consultations, 583 they expected the doctor to take full control while they remained passive as they were acclimated to 584 in Turkey. Some ethnic minorities especially immigrants from non -western cultures, may have a 585 preference of a certain power dynamic in the doctor-patient relationship. They may prefer the 586 doctor to assume a dominant role and take full control of the consultation, and do not expect nor 587 desire to be involved in decisions about the treatment as they expected the doctor to know 588 everything and dictate the course of action. This expected doctor -patient relationship is not in line 589 with practices in developed western countries where doctors are encouraged to involve the patient 590 as much as possible while determining the course of treatment. We had hypothesized that some 591 ethnic minorities may rate the overall satisfaction more negatively if they felt pressured to get more 592 involved in the consultation process as they may not be accustomed to it. When the (Involved in 593 decisions about treatment) variable was added to the multiple regression analysis, the change on 594 effect of ethnicity on overall satisfaction was minimal. 612 Multilevel regression analysis identified a number of factors relating to service provided at GP 613 surgery that explain the observed ethnic disparities. Significant factors included: ease of using GP 614 website; frequency of seeing preferred GP; being treated with care and concern; confidence of . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted June 21, 2022. ; https://doi.org/10.1101/2022.06.20.22276629 doi: medRxiv preprint