Measuring and valuing broader impacts in public health: Development of a sanitation‐related quality of life instrument in Maputo, Mozambique

1 INTRODUCTION

Nearly two billion people globally lack access to basic sanitation (WHO & UNICEF, 2021). This failure to separate human excreta from human contact represents an enormous public health challenge. Every year 26 million disability-adjusted life years (DALYs), including 430,000 child deaths from diarrhoea, are attributable to inadequate sanitation (Prüss-Ustün et al., 2019). From a health system perspective, averting this disease burden would be highly valuable. Toilet users, however, often identify benefits of improved sanitation for privacy, safety, reduced smells, and dignity, as more important than reducing infectious disease (Elmendorf & Buckles, 1980; Jenkins & Curtis, 2005; Mukherjee, 2001). These are attributes of good quality of life (QoL), and contribute to health in its broadest sense including mental and social wellbeing (WHO, 1948). Therefore, the QoL benefits beyond infectious disease are likely to underpin household willingness to pay (WTP) for sanitation improvements, with consequences for their measurement and valuation in cost-benefit analysis (CBA).

Studies have identified associations between better sanitation and general measures of mental wellbeing (Caruso et al., 2018; Gruebner et al., 2012). However, the QoL outcomes such as privacy and safety are not captured in DALYs. Quality-adjusted life years (QALYs) are less-used in lower- and middle-income countries (LMICs), and have never been used to evaluate sanitation interventions. In any case, QALYs would also be unlikely to capture benefits of sanitation beyond morbidity and mortality from infectious disease. This is because changes in the QoL outcomes prioritised by toilet users would not be picked up by measures widely used to weight QALYs such as the EQ-5D, with the exception of consequences for anxiety/depression (Brazier et al., 2016; Euroqol Group, 2009).

Despite the importance of outcomes such as privacy and safety to users, economic evaluations of sanitation interventions have never managed to specifically include them, in the absence of means for their measurement and valuation (Hutton & Chase, 2016). The only way in which sanitation-related QoL outcomes have been valued economically is through WTP for toilets (Hutton & Chase, 2016). Willingness to pay for benefit valuation is often considered dissatisfactory in a health context due to the biases it introduces (Cookson, 2003), particularly from ability to pay (Coast, Smith, & Lorgelly, 2008). The bias is likely to be toward under-valuation if a respondent's WTP for a toilet is taken to comprise the total benefits of sanitation (e.g., morbidity, mortality, healthcare savings, time savings, privacy, safety, etc.). Accurate estimation is more likely if these benefits are estimated separately and aggregated, as is preferred in CBA (Boardman et al., 2018), and in practice all sanitation CBA studies do this (Hutton & Chase, 2016).

No quantitative measure capturing sanitation interventions' QoL outcomes with a user-derived valuation scheme exists (Sclar et al., 2018). The only standalone measure of such outcomes is the women's sanitation insecurity profile, which comprises 60 survey questions spread over seven factor scales measuring a mixture of practices, experiences and feelings (Caruso et al., 2017). While it is appropriate for non-economic purposes, the women's sanitation insecurity profile's application of equal weighting and lack of an overall score preclude its use in economic evaluation (Brazier et al., 2016). It was also developed with (and designed for) women only, while sanitation affects the general population. To inform the allocation of public funds, a respondent-weighted index with a single overall score designed for use in the general population is needed.

The lack of an outcome measure applicable for use in economic evaluation may be contributing to misallocation of the US$ 20 billion invested annually in sanitation in LMICs (WHO, 2017). Health economists are increasingly using capability-based outcome measures as a way to broaden the evaluative space beyond the value of health alone (Coast, Smith, & Lorgelly, 2008; Greco et al., 2016). The capability approach focuses on the value of what people are able to be and do (Sen, 1980, 1993) and is well-suited to a multi-dimensional conceptualisation of QoL (Coast et al., 2015). Many capability-based measures have been developed for different economic evaluation purposes both in high-income and low-income settings (Coast, 2019; Greco, 2016; Simon et al., 2013).

In Mozambique, where our study took place, 39% of the urban population (77% in rural) do not have basic sanitation, defined as “improved” facilities which are not shared with other households (WHO & UNICEF, 2021). A third of those use a shared improved facility. Urban sanitation services are a municipal responsibility and each city has policies and strategies in place (CMM, 2017; MOPH, 2011), but access to basic sanitation areas is increasing at only 1.5 percentage points per year (WHO & UNICEF, 2021).

In this study, we aim to develop and value a measure of sanitation-related quality of life (SanQoL) in urban Mozambique, based on prior qualitative research in the setting (Ross et al., 2021). The underlying objective was to enable classification of “sanitation states” and their valuation, in support of economic evaluations of sanitation interventions. We include exploration of the measure's validity and reliability. The study was nested within the Maputo Sanitation (MapSan) trial (clinicaltrials.gov registration: NCT02362932), which evaluated the impact of a shared urban sanitation intervention on children's enteric infections (Knee et al., 2021).

2 METHODS 2.1 Study setting

We undertook this study in Maputo, Mozambique, in the low-income neighbourhoods of the Nhlamankulu district where multi-household compounds with shared sanitation facilities are common (Brown et al., 2015). Of Maputo City's 1.1 million population, 70% live in informal settlements (INE, 2019; UN-HABITAT, 2010). Approximately 89% use non-sewered sanitation facilities (Hawkins & Muximpua, 2015). In the control group of the MapSan trial, households used low-quality shared pit latrines, which are common in low-income areas. These latrines typically comprised unlined pits with squatting slabs made of wood or car tyres, and no water seal (u-bend) to provide a barrier against smells and flies. Few pit latrines have roofs, and the walls are often made with sections of scrap corrugated iron or plastic sheeting, with makeshift fabric doors. Together, these conditions lead to unpleasant, unsafe toilets providing little privacy.

In the intervention group, an international non-government organisation provided a subsidised pour-flush toilet with concrete superstructure, discharging to a septic tank. The intervention toilets had two design types, depending on user numbers. The first was a shared toilet with one stance (cubicle) to be used by around 15 people, at 85% subsidy. The second was a Community Sanitation Block with two stances, to be used by a minimum of 21 people, at 90% subsidy. Both STs and CSBs were built from concrete blocks, with metal doors lockable from the inside. More information on the setting, including photos of toilet types, is provided in Supplementary Material A. Our study received approval from the Comité Nacional de Bioética para a Saúde (IRB00002657) at the Ministry of Health in Mozambique and the ethics committee at the London School of Hygiene & Tropical Medicine (Ref: 14609). Informed, written consent was obtained from all participants. Deidentified individual participant data, codebook, and replication code are available online (Ross, 2021).

2.2 Overall study design

We followed measure development methods common in health economics for the purpose of economic evaluation (Goodwin & Green, 2016), whereby the final measure comprises index values anchored at 0 and 1 (Drummond et al., 2015). The descriptive system comprises a set of items with categorical response scales, each representing one dimension of the construct being measured (Brazier et al., 2016). Items are selected primarily for content validity (Fayers & Machin, 2015), and an individual's combination of responses under the descriptive system comprises their “health state” (Brazier et al., 2016) or “capability state” (Coast, Flynn, et al., 2008). Methods such as factor analysis are not applied, and instead states are valued as an index using preference elicitation.

We adapted these methods to develop a measure of SanQoL. The capability approach to welfare economics (Sen, 1980, 1993) comprised our overarching theoretical approach to defining and measuring QoL, which informed both qualitative and quantitative methods. The target population for our measure is people living in urban settings where poor sanitation is common. The primary intended use is in economic evaluation of sanitation programmes. This required the measure to have a small number of items, since trading-off more than seven attribute levels simultaneously in preference elicitation tasks is generally considered too cognitively demanding (Hensher et al., 2015; Mangham et al., 2009).

Initial qualitative research in the same Maputo setting defined the construct to be measured as “the subset of overall QoL which is directly affected by sanitation practices or services” (Ross et al., 2021). This draws on common definitions of health-related QoL (HRQoL), and is aligned with the third of the four types identified by Karimi and Brazier (2016) in their review of definitions of health and HRQoL. Specifically we build on how Peasgood et al. (2014, p. 17) describe narrower definitions of HRQoL: “the sub-set of the important or most common ways in which health or health care impact upon well-being.” The scope of sanitation practices is as perceived by users but is assumed to extend beyond defecation and urination to include for example menstrual hygiene, as well as any related practices users consider important, such as bathing. The capability approach was reflected in the design of the qualitative study (Ross et al., 2021). For example, the topic guides focused on a broad evaluative space (“a good life”), we used focus groups to engender the deliberation encouraged by the capability approach, and we triangulated findings on attributes' relative importance using methods from cognitive anthropology (Weller & Romney, 1988).

The qualitative research resulted in a conceptual model (Figure 1) centred on five attributes: disgust, health, privacy, safety and shame. These were the attributes toilet users in this setting considered most important, and are largely consistent with the broader qualitative literature in other countries and settings (Novotný et al., 2018). Some studies have identified convenience or water supply, for example, as important (Jenkins & Curtis, 2005; Sahoo et al., 2015). However, these were not raised by our participants, reflecting that the social and environmental context in which a toilet “commodity” (Sen, 1985) is used affects the capabilities an individual can derive from it (Ross et al., 2021).

image

Conceptual model for sanitation-related quality of life (Ross et al., 2021)

In this paper, we present the subsequent stages of measure development (Table 1), adhering to the standards of the International Society for QoL Research for patient-reported outcome measures (Supplementary Material B) (Reeve et al., 2013). We conducted all analyses in Stata 16 (StataCorp, 2019).

TABLE 1. Stages followed in measure development Stage Objective Method Main output 1. Conceptual model To identify what people most value about sanitation, using qualitative research In-depth interviews (n = 19) and focus groups (n = 8) Conceptual model 2. Piloting To ensure items are understood, and identify floor/ceiling effects Pilot interviews (n = 64) and cognitive interviews (n = 28) Five-item descriptive system 3. Valuation To estimate the relative value of attributes Rank sum method (n = 424) with a view to subsequent discrete choice experiment SanQoL index values anchored at 0 and 1 4. Validity & reliability To assess the extent to which the instrument measures what it intends to measure (validity) and does so consistently (reliability) Quantitative survey amongst intervention and control groups of a trial (n = 424) Evidence of construct validity, convergent validity, and test-retest reliability. Abbreviation: SanQOL, sanitation-related quality of life. 2.3 Item development and piloting

We developed several possible items for each of the five attributes identified by the qualitative research (Streiner et al., 2015), aiming to focus attention on capabilities by framing questions with “can you…?” or “are you able to…?” A long-list of 36 items was reviewed internally, then by 14 external experts. We undertook piloting and cognitive interviewing during April 2019, recruiting a team of four enumerators fluent in Portuguese and Changana, the first and second most commonly spoken languages in Maputo. One author (Z. A.), a Mozambican fluent in Portuguese and English, translated items into Portuguese. (I. R.), an intermediate Portuguese speaker, extensively discussed meanings with (Z. A.) and the enumerator team. We piloted items with 64 individuals from the target population, and undertook further cognitive interviewing with 28 of the pilot respondents (Bowden et al., 2002). In cognitive interviews, after each SanQoL item, the respondent was asked to explain back the question in their own words and discuss the ease of understanding the question. Enumerators rated their explanation on a scale of full, partial or no comprehension. For each of the five attributes, we identified one item which best achieved face and content validity, considering the qualitative findings and cognitive interviews. These five items comprise the descriptive system, with responses on a four-level ordinal frequency scale: always, sometimes, rarely, never (Table 2, and in Portuguese in Supplementary Material C). The descriptive system therefore contains 1024 (=45) combinations of attribute levels, or “sanitation capability states.” Items were framed as direct questions such that “always” was consistently the best outcome.

TABLE 2. Descriptive system for the SanQoL measure Attribute Questionnaire item Responsesa Disgust Can you use the toilet without feeling disgusted?

Always

Sometimes

Rarely

Never

Health Can you use the toilet without worrying that it spreads diseases? Privacy Can you use the toilet in private, without being seen? Shame Can you use the toilet without feeling ashamed for any reason? Safety Are you able to feel safe while using the toilet? Abbreviation: SanQoL, sanitation-related quality of life. a respondents can choose “prefer not to answer” for any item. The recall period was 4 weeks (Supplementary Material C). 2.4 Valuation

Valuation aims to aggregate item responses into a single score, weighted by the relative value of attributes elicited from the target population. We undertook a quantitative survey in May 2019 with the same enumerator team, using the mWater (2019) application on smartphones. We aimed to recruit at least 400 people aged over 18 living on trial-enrolled compounds, stratified by intervention status and gender. This meant a range of outcomes was likely, due to the diversity of low-quality toilet types in control compounds and the high-quality toilets in use in intervention compounds. We recruited a maximum of two people (one man, one woman) from each compound, on condition that they were not from the same household. In addition to SanQoL items, we collected data on water supply and sanitation services, demographic characteristics, and calculated an asset-based wealth index following standard methods (Vyas & Kumaranayake, 2006). Finally, we asked respondents to rank the five SanQoL attributes on a visual analogue scale, according to their relative importance (Supplementary Material D).

With a view to subsequent valuation by discrete choice experiment, our interim valuation strategy was based on the rank sum method (Stillwell et al., 1981), following previous measures (de Kruijk & Rutten, 2007; Greco, 2016). This required scoring attributes on a 0-3 scale based on item responses (0 = never, 1 = rarely, 2 = sometimes, 3 = always), and estimating attribute weights from the rank data (Equation 1). We calculated SanQoL index values by combining attribute scores and weights, then rescaling to anchor at 0 and 1 (Equation 2). Zero represents “no sanitation capability” and one represents “full sanitation capability,” building on the Investigating Choice Experiments and Capabilities measures' anchor points (Coast, Flynn, et al., 2008).

Equation 1 – attribute weights for a population: urn:x-wiley:10579230:media:hec4462:hec4462-math-0001(1) Equation 2 – SanQoL index value for an individual: urn:x-wiley:10579230:media:hec4462:hec4462-math-0002(2)where: urn:x-wiley:10579230:media:hec4462:hec4462-math-0003 is the weight of the ith attribute; urn:x-wiley:10579230:media:hec4462:hec4462-math-0004 is the number of attributes; urn:x-wiley:10579230:media:hec4462:hec4462-math-0005 is the mean rank of the ith attribute in the population; urn:x-wiley:10579230:media:hec4462:hec4462-math-0006 are attribute scores ranging from 0 to 3 for the jth individual, where “always” = 3 and “never” = 0; urn:x-wiley:10579230:media:hec4462:hec4462-math-0007 is the SanQoL index value for the jth individual.

We explored differences in attribute ranks by gender, whether the respondent was elderly (aged 60 or over), and treatment group. The rationale for including gender was that women experience sanitation differently to men in that they squat for urination, manage menstrual hygiene, and are more likely to fear and experience “peeping” or assault (Tilley et al., 2013). The rationale for including elderliness was that older people are particularly impacted by poor sanitation, predominantly as a result of disabilities that occur with ageing (Groce et al., 2011). In Mozambique, an elderly person is defined in law as anyone aged 60 or older (Castel-Branco & Andrés, 2019). The rationale for including treatment group was to assess whether people who had been using higher-quality toilets for four years (intervention group) had different perceptions of the relative value of the attributes. Rank is an ordered categorical variable, so we used mixed-effects ordered logit models, clustering standard errors at the compound level. We regressed on the rank for each attribute (which ranged from 1 to 5), including as covariates gender, aged 60 or over, and treatment. We also included as covariates any variables which both (i) differed at the 5% level along any of those three lines; and (ii) plausibly influenced ranking. The two variables meeting those criteria were the wealth index (differed by treatment status) and primary education completion (differed by gender and by aged 60 or over). P-values less than 0.05 were considered statistically significant evidence of association. As a robustness check, we also used generalised linear mixed models (GLMM) to analyse rank as a continuous variable, with the same set of covariates.

2.5 Validity and reliability assessment

We assessed internal reliability using item-total correlation and Cronbach's alpha, for which common acceptability thresholds are 0.4 (Ware et al., 1980) and 0.7 (Nunnally, 1978) respectively. There is debate about whether these metrics are appropriate for indices, since each item measures a different dimension (Konerding, 2013), but we include them because our measure comprises a single overall score to represent a single construct. We examined distributions of frequency endorsements in aggregate and by gender (Terwee et al., 2007). We assessed test-retest reliability by re-interviewing 69 respondents (16% of sample) two weeks after the original interview (Streiner et al., 2015). We used a two-way mixed effects model (Koo & Li, 2016) to evaluate the intraclass correlation coefficient (ICC) for the SanQoL index value, against an acceptability threshold of 0.7 (Terwee et al., 2007).

To assess construct validity, we pre-specified hypotheses (Table 3) about the presence or absence and direction of associations between each attribute score and a set of user and toilet characteristics. We tested these using GLMM, because attribute scores represent points on an underlying continuous scale, with standard errors clustered at the compound level. We regressed on each attribute score, including as covariates: gender; aged 60 or over; wealth index; primary education completion; whether the toilet had a concrete or tile floor; whether the toilet had masonry or zinc sheet walls; whether the toilet had an inside lock; and, whether the enumerator smelt faeces on entering toilet. The hypotheses drew on the previous qualitative work (Ross et al., 2021) and the broader literature on motives for sanitation behaviours and mental wellbeing (Novotný et al., 2018; Sclar et al., 2018).

TABLE 3. Hypothesised presence or absence of associations Variable Association hypothesised Rationale for hypothesis Disgust Health Shame Safety Privacy User characteristics Female No No No Yes (−ve) Yes (−ve) Women might have higher acceptability thresholds for safety and privacy, since they are at higher risk of peeping, sexual harassment, and assault. Aged 60+ No No No No No No reason to expect any individual item to systematically covary with being elderly. Wealth index (continuous) No No No No No No reason to expect any individual item to systematically covary with wealth. Toilet characteristics Toilet floor material (high/low quality) Yes (+ve) Yes (+ve) Yes (+ve) Yes (+ve) No The quality of the toilet floor might affect all attributes, but there is no obvious rationale for privacy. Toilet wall material (high/low quality) No No Yes (+ve) Yes (+ve) Yes (+ve) The quality of the toilet wall might affect the extent to which it prevents others from seeing toilet users, thereby affecting privacy, safety and shame. Toilet locks from the inside No No No Yes (+ve) Yes (+ve) An inside lock might directly improve privacy and safety, but there is no obvious rationale for disgust, health and shame. Enumerator smells faeces on entry Yes (−ve) Yes (−ve) Yes (−ve) No No A smelly toilet is likely to affect disgust, shame and perception of health risk. There is no obvious rationale for safety and privacy. Note: “Yes (−ve)” means that having the specified characteristic is hypothesised to be associated with lower (worse) scores for the attribute, for example, women are hypothesised to have lower privacy scores than men, all else being equal. “Yes (+ve)” indicates a hypothesis of higher scores and “no” of no significant association.

We assessed convergent validity by correlation (Pearson's r) between SanQoL index values and the WHO-5 mental wellbeing index (Topp et al., 2015). We expected correlation to be positive but less than 0.5, since sanitation is unlikely to be a primary driver of mental wellbeing (Sclar et al., 2018). Finally, we investigated the convergence of SanQoL index values between respondents using the same toilet. We used inter-rater methods to test the hypothesis that responses would be positively correlated but not equal, because any two people may experience the same toilet differently. We calculated the ICC using a one-way random effects model (Koo & Li, 2016). Interpretation of this ICC is “fair” (0.40–0.59), “good” (0.60–0.74), or “excellent” (>0.75) (Cicchetti, 1994).

3 RESULTS 3.1 Piloting and cognitive interviews

Of the 64 piloting respondents, 48% were male and 52% female. No single category per item received greater than 50% of endorsements in piloting, well-below an 80% benchmark (Streiner et al., 2015). As is desirable, there were no floor or ceiling effects, that is, absence of high proportion of respondents with maximum/minimum scores (Streiner et al., 2015). Cognitive interviewing showed that items could be understood, and no changes to the English descriptive system were required. However, there was one change to the Portuguese. The disgust item (Table 2), originally framed as an adjective (enojado = disgusted) was reframed as a noun (nojo = disgust), considered more natural in the Portuguese spoken in the setting.

3.2 Valuation

We interviewed 424 individuals (220 female, 204 male) each from a different household, with a response rate of 99%. Respondent characteristics are tabulated by gender in Table 4 (and by trial arm in Supplementary Material E). Respondents lived on 275 MapSan trial compounds (131 control, 144 intervention). About two thirds had completed primary education, with slightly more men (70%) than women (57%) having done so. There was near-universal access to piped water connections (98%). The vast majority (89%) of respondents shared their toilet with other households, with a mean of 12.2 people sharing each toilet stance (cubicle).

TABLE 4. Characteristics of quantitative sample, overall and by gender Overall (n = 424) Male (n = 204) Female (n = 220) Respondent demographic characteristics Respondent age 39.9 (15.3) 39.3 (15.1) 40.5 (15.5) Respondent has a partner 214 (50%) 116 (57%) 98 (45%) Household size 5.1 (3.0) 4.7 (3.0) 5.5 (2.9) Number of children under-14 1.3 (1.6) 1.1 (1.3) 1.4 (1.7) Other respondent characteristics Completed primary school or above 268 (63%) 143 (70%) 125 (57%) Completed secondary school or above 51 (12%) 30 (15%) 21 (10%) Moderate problems walking about, or worse 25 (6%) 9 (4%) 16 (7%) Moderate pain or discomfort, or worse 38 (9%) 12 (6%) 26 (12%) Respondent housing Dwelling has cement or tiled floor 394 (93%) 191 (94%) 203 (92%) Dwelling has concrete exterior walls 283 (67%) 131 (64%) 152 (69%) Dwelling has zinc or concrete roof 424 (100%) 204 (100%) 220 (100%) Household has access to electricity connection 359 (85%) 175 (86%) 184 (84%) Rents dwelling 114 (27%) 56 (27%) 58 (26%) Compound-level water and sanitation characteristics Piped water connection 416 (98%) 199 (98%) 217 (99%) Water available at least 8 h/day 209 (49%) 101 (50%) 108 (49%) Uses on-plot toilet 416 (98%) 201 (99%) 215 (98%) Pour-flush to septic tank (intervention) 222 (52%) 103 (51%) 119 (54%) Pit latrine (control) 202 (48%) 101 (49%) 101 (46%) Shares toilet with other household(s) 377 (89%) 186 (91%) 191 (87%) Number of households sharing stance 3.2 (1.6) 3.3 (1.6) 3.2 (1.7) Number of people sharing stance 12.2 (

留言 (0)

沒有登入
gif