I’m currently attending this large conference in lovely Toronto and trying to pack-in as many sessions as possible. A handful of papers have stood out to me – two evaluations of on-going pay-for-performance schemes in health and two methodological papers related to the economics of obesity. Though I am excited to share them with readers, we need to remember that these are all works in progress to varying degrees – most of them not yet even working papers – and so all reported findings should be taken with the caveat that they are subject to change.
Tisemarie Sherry, Sebastian Bauhoff, and Manoj Mohanan leverage two DHS surveys from Rwanda to evaluate the Rwandan Results Based Financing Scheme. A counter-part paper by Paulin Basinga et al. (recently published in the Lancet) use dedicated facility and household survey data to evaluate the same program and find significant gains to the highest incentivized indicators such as in-facility deliveries, but little change to other targeted indicators. This conference paper by Sherry and co-authors is a complementary approach that uses separate national data representative of women of child-bearing age. The findings are broadly consistent with Basinga and friends. Sherry et al. find a significant improvement in the in-facility delivery rate and in one select measure of the quality of antenatal care (urinalysis during an antenatal visit) but little observable change in the other indicators contained in the DHS.
The authors also investigate the possibility of substitution away from non-incentivized activities behavior among providers and find no change in the rates of non-incentivized indicators that were captured by the DHS. Importantly, the DHS not only measures health service utilization but also contains select measures of health outcomes such as the two-week prevalence of the common child illnesses diarrhea and fever.
Interestingly the study finds no impact of RBF on these health outcome measures although I wonder how responsive we expect these particular measures to be given the structure of the RBF intervention. For example, many determinants of communicable disease prevalence are not specifically targeted by RBF. Also the increased number of child visits to facilities as incentivized by the RBF may itself induce greater rates of self-reported illness. I am curious to see analysis that instead looks at stock measures of health such as anthropometry – these indicators can be assessed by third-party and would presumably reflect the cumulative exposure of the child to the program. Nevertheless I think this paper will be another important link in the growing evidence surrounding pay-for-performance (P4P) in health.
John Peabody, Stella Quimbo, and co-authors contrast a pay-for-performance scheme at hospitals with expanded inpatient health insurance in the QIDS study in the central Philippines. This is a three arm randomized control trial where the hospital is ether introduced to a P4P scheme or were given funds to introduce a universal coverage insurance scheme. The study is confined to 30 hospitals, including controls, and so to maximize power the authors institute a triplet-matched randomized assignation scheme
The introduction of insurance coverage results in significant decreases relative to controls in both wasting and in c-reactive protein (a measure of illness induced inflammation) among discharged patients. Why might this occur? The authors posit as a transmission channel the greater likelihood of completion of pharmaceutical treatment as well as higher food expenditures as a result of the financial protection afforded by the insurance scheme. Under the P4P scheme, which paid quality-of-service bonuses to the hospital staff on the order of 5% of total salary, there were also reductions in wasting among the discharged child patients as well as gains in mother-reported child health. The authors’ stated next steps are to assess the relative performance of the two interventions against each other and to refine the comparative cost-effectiveness analysis. Right now it appears that the P4P scheme is substantially more cost-effective in reducing child wasting after discharge.
I was also drawn to several sessions on obesity – a pervasive health problem in rich countries and a growing one in numerous developing countries. I particularly enjoyed a working paper by Mary Burke and Frank Hailand that explores the validity of self-reported food consumption and caloric intake in the US. The motivating problem is that self-reported caloric intake and physical activity have an unexpectedly weak relationship to observed BMI and obesity in the multi-wave US health and nutrition survey NHANES.
The authors suggest that one reason for this attenuation is non-classical measurement error where underreports of caloric intake increase as true caloric intake goes up, perhaps in part due to social desirability concerns. It also appears that respondents may overstate true physical activity levels again because of social desirability concerns. The authors exploit metabolic energy accounting – at a steady-state BMI, the quantity of energy intake must exactly match the energy expenditure – to derive bounds on the self-reported values.
The authors also include in the analysis a proxy – the ratio of self-reported caloric intake relative to self-reported activity levels – for the possible joint measurement error in self-reported behaviors. In the population, this ratio is less than one, suggesting that the average respondent underreports caloric consumption relative to physical activity. Applying these two “fixes” to the data results in a much more pronounced and expected relation between reported food consumption, reported physical activity, and BMI.
Finally, Euna Han and Lisa Powell exploit US cross-state variation in restaurant and fast-food taxes to explore the association between the relative price of fast food and population obesity. The proportion of spending in the U.S. on food outside the home has more than doubled from 20% to 41% over the 1960-2008, and outside food is often higher in calories. The authors supplement NHANES data with trade data on fast-food and other prices. Their preferred fast-food price measure is a weighted index of cheeseburgers, pizza, and fried chicken (I think it is safe to assume commodity homogeneity across standardized fast-food outlets in different states so I am not worried about the possible confounder of quality variation when using spatial price differences to estimate a relationship).
Interestingly the authors find no association between the relative cost of fast-food and mean or median BMI (the body-mass index, a standard measure of obesity). However, using quantile regressions, they do find a significant association at the upper tails of BMI in the population – the 75th and 90th percentiles of BMI are significantly higher in areas with lower cost fast-food, and this is especially true when individuals with large BMI also have children. Yes, more evidence on the health benefits of taxing fast-food or soda, but this also speaks to equity considerations since fast-food is valued by segments of the population due to its relative convenience.