Toward the development of a short multi‐country person‐centered maternity care scale

To develop a shortened, valid and reliable scale applicable across multiple settings for routine monitoring of person‐centered maternity care (PCMC).

person-centered care. [8][9][10] The concepts of respectful maternity care (RMC) are incorporated in PCMC as part of the broader interest in person-centered care, 11,12 and capture the experience dimensions in the WHO vision for quality of maternal and newborn health. 7 While there is growing consensus on the importance of PCMC, much work is needed to operationalize it in developing settings. A recent systematic review found 36 instruments in the literature for measuring women's birth experiences (as of January 2016). Only seven of the tools had psychometric properties indicative of high quality scales, and none were validated in an LMIC; furthermore, only three were sufficiently broad to holistically measure PCMC. 13 Since this review, two scales have been published for measuring women's birth experiences in LMICs, both demonstrating high validity and reliability: the 15-item RMC perception scale by Sheferaw et al. 14  scale represents a broader measure of women's experience than the RMC perception scale and uses a response format that proved easier to interpret in cognitive interviews. 9 The PCMC scale was rigorously developed, including a thorough literature review, expert reviews and cognitive interviews, yielding strong content validity. It has been applied in urban and rural contexts in Kenya, India and Ghana, increasing its generalizability for use in LMICs, compared to existing tools; however, given its length, one limitation of the PCMC scale is its applicability to program settings-it was developed to be comprehensive, informing its large number of items.
While this is useful for research and comprehensive needs assessment, program implementers tend to value shorter, easier-to-implement, measurement tools. Additionally, because of the context-specific nature of certain PCMC dimensions, only a subset of the items may be applicable to most settings. To ensure that not only was PCMC measurable-but that PCMC was measured routinely-a scale was required that was easier to administer and interpret. For these reasons, the aim of the present study was to develop a shorter, more simplified PCMC tool that could be applied by program implementers across multiple settings.  Exploratory Factor Analysis (EFA) was then conducted using splithalf reliability datasets for each setting. EFA was an iterative process replicated consistently across settings. Because the domains of PCMC are theoretically related, oblique rotation was used to allow for correlations between the rotated factors. Items that loaded with a uniqueness value above 0.80 or loaded below 0.30 on any factor were removed. Efforts were made to retain highly regarded items by lowering this threshold to 0.10 for items with more than 80% expert backing. The shortened setting-specific scales were tested in the remaining sample, using confirmatory factor analysis (CFA).

| MATERIALS AND METHODS
To identify a core scale in accordance with the study objective, items retained throughout the process of scale shortening in any two of the three samples were included in a final set of questions representing a potential multiple-setting PCMC scale. A final round of EFA was conducted using the principal factoring method and oblique rotation to identify the underlying factor structure and examined Cronbach's alpha for each of the resulting factors. Confirmatory factor analysis was performed to identify the goodness-of-fit of the core scale in each setting, allowing error terms to correlate between items within the same proposed domain. Goodness-of-fit was estimated using root mean square error of approximation (RMSEA) and the comparative fit index (CFI). Criterion validity of the shortened scale was assessed by summing responses on the items in the final scale to generate scores, which were regressed on measures of satisfaction with maternity services. Finally, intraclass correlation (ICC) analysis was conducted to examine the relationship between scores derived from the short version and the original 30-item scale in each setting. P<0.05 was considered statistically significant.  Ninety-six responses were received to the online survey of global MCH experts between September and November 2017. Table S1 shows the characteristics of the experts surveyed: 86% of respondents were women, with 13 years' MCH experience on average. In Kenya, ordered logistic regression was used to determine if PCMC scores can predict patient-reported levels of satisfaction and willingness to deliver in the same place again. In India, logistic regression was used to determine if PCMC scores can predict willingness to deliver in the same facility. In Ghana, all women indicated high levels of satisfaction, so the analysis was not conducted. An increase in PCMC scores (as measured by the 13-item reduced scale) was associated with an increase in the odds of reporting higher general satisfaction in Kenya, and with willingness to deliver in the same place again in both Kenya and in India. b P<0.05 was considered statistically significant.

Surveys
T A B L E 3 Reliability, distribution, and goodness-of-fit statistics for the 13- Abbreviations: CFI, comparative fit index; Min, Minimum; Max, Maximum; RMSEA, root mean square error of approximation. a The possible range of scores on the 13-item unidimensional scale is 0 to 39. To better understand goodness-of-fit, the scale structures identified in each setting were tested across the three settings. The 2-factor structure identified in India provided the best fit, and only those statistics are reported here for clarity. Error terms were allowed to correlate between items within the same proposed domain.

| DISCUSSION
The present study used a data-driven approach to shorten the 30-item PCMC scale developed in Kenya by 47% to 13 items while maintaining high validity and reliability. 11 The EFA indicated high construct validity, although it presented both a unidimensional and two-dimensional factor structure, both of which fit the data in all three countries. Given that EFA can produce excessive factors with ordinal items, the authors were confident that the 13 items represented a unidimensional PCMC scale that could be applied across the settings. 17 Hypothesis testing for convergent validity in the India and Kenya samples suggested higher PCMC on the 13-item scale was associated with the willingness of women to deliver in the facility again in Kenya and India, and with increased satisfaction in Kenya.
Person-centered maternity care is a healthcare related patientreported outcome, akin to measures of patient experience used for other in-patient and out-patient needs. 18,19 Minimum standards for patient-reported outcome measures provided a systematic approach to assess the shortened PCMC measure. 18,20 The COSMIN  T A B L E 5 (Continued) have different preferences. 24 However, the authors recognized the importance of these items and further work to refine the short PCMC scale will consider ways in which to prioritize their inclusion.
Items were excluded which were related to the facility environment based on the initial criteria to remove items that might be beyond the capacity of individual providers to change. Also, the experts whose opinions informed the first iteration of scale shortening did not consider the questions related to water, electricity, waiting time, and crowding to be priorities in assessing PCMC, so these items were excluded in the shortened scale. This may have reflected an assumption that infrastructural inputs were foundational requirements in care settings. Many experts were involved in research or program implementation associated with international efforts or organizations, so their own experiences as to a "normal" baseline of care may have influenced the exclusion of more basic inputs to the provision of care. The full 30-item PCMC scale included all of these items and presented a more comprehensive alternative to the shortened scale. Second, the data for these analyses were based on convenience samples within facilities and health units, and so the findings for each setting was not generalizable to the whole country or even the particular setting. Selection bias was also a potential limitation of the non-random sample.
Despite these limitations, this shortened PCMC scale showed promise in being able to routinely assess the experience of care that women receive during childbirth. With only 13 items, program managers could easily determine which items receive low scores, and work to make improvements on aspects of care within their control.
The generalizability of the scale needs to be assessed in more representative samples and in additional settings, including Latin American and Southeast Asian populations, and with more women receiving care in the private sector. Finally, data from a wider range of women and at multiple time points will help to determine if the measure is responsive and actionable, and will allow for greater interpretability of scores, particularly when correlated with other clinical and patient-reported outcomes.

AUTHOR CONTRIBUTIONS
PA and KF contributed equally to this manuscript. PA contributed to the conception and design of the study, data collection, and data analysis. KF contributed to the design of the study and data analysis.
MS contributed to the design of the study and data collection. RA contributed to data collection. DM contributed to the design of the study. NC contributed to the design of the study and data analysis.

SUPPORTING INFORMATION
Additional supporting information may be found online in the Supporting Information section at the end of the article. Table S1. Distribution of Respondents in MCH Expert Survey. Table S2. Items in the reduced scales generated in each setting.