We used patient data collected before, during, and at discharge from hospital treatment to assess the primary effects of the Gemura hospital feeding program on nutritional status, and secondary effects on recovery, patient costs of care, patient satisfaction, and health-related quality of life.

Our findings on food intake, weight change, and BMI show that there were statistically significant differences between the intervention and control groups, indicating the program had a positive effect on nutrition. Since the proportion of Gemura group patients reporting having eaten enough while in hospital was almost three times (2.96) that of the control group, the intervention clearly improved food intake.

Results also suggest that the intervention had an effect of mitigating weight loss. Mean weight change was negative (-1.5%) among the control group with as much as 36.6% losing more than 2% of baseline body weight, whereas weight change was positive (+0.6%) in the Gemura group with only 20.4% losing more than 2% body weight. Considering that the control group had a much shorter average length of stay (4.81 days) than the Gemura group (8.35 days), the difference in weight change is dramatic as the loss in the control group occurred in almost half the time. It is conceivable that the control group could have experienced even greater weight loss with longer lengths of stay under similar conditions, with more approaching the generally considered clinically significant 5% body weight loss. Fortunately, weight loss even in the control group was not as severe as other observational findings in Sub-Saharan Africa, namely one multicentre study in South Africa, Kenya, and Ghana with 20.6% of patients having lost more than 5% body weight in hospital,³⁵ and another study in Benin in which two-thirds of the sample lost more than 2%.³⁶ Comparable hospital weight change findings in Rwanda are not available, so closer context-specific conclusions are more difficult. These findings of relatively small change (most lost less than 2%) in weight over an average length of stay not more than one week in the entire sample are consistent with well documented research showing that body weight is not highly sensitive to large fluctuations in energy intake over the short-term.³⁷

Positive nutrition effects of the intervention are further supported by the improved BMI distribution in the Gemura group as compared to the control group. Though both groups at baseline had similar BMI distribution, with higher prevalence of malnutrition in the intervention group, the Gemura group experienced a decrease in malnourished, an increase in normal weight, and a decrease in combined overweight and obese, indicating convergence toward normal.

Conversely, the control group showed an unfavourable trend of decreasing weight across all categories except combined overweight and obese. Most notably, since study arm was strongly associated with the prevalence of malnutrition, the results indicate that the intervention mitigated hospital malnutrition.

The difference-in-difference analysis however, found little evidence of a significant relationship between the feeding intervention and the weight and BMI changes. The analysis did show that the program had a higher magnitude of effect than other variables on BMI, but overall indicated that there may have been other, unmeasured variables that contributed to the BMI and weight changes. DID analysis may be limited in capturing the nuance of the distribution effects of BMI and weight changes, where convergence toward normal is more desirable than one direction effects.

It was hypothesized that observed improvements in nutrition would lead to better outcomes in patient recovery, as measured by shorter hospital length of stay. Our findings on LOS and patient reported recovery provided limited evidence to support this hypothesis, since LOS was longer in the intervention group than the control group and patient reported recovery was similarly favourable in both groups. Indeed, LOS results from the intervention group were quite unexpected, since secondary routine data on average LOS across hospital wards in the intervention hospitals during the study period indicate LOS much closer to the control hospital results in our study. One reason for the difference could be that our study sample purposively included patients assessed as socioeconomically vulnerable by hospital social workers.

Because LOS had such weak association with study arm, with less observed effect than all other variables we tested including age, disease category, baseline BMI, province, and sex, the LOS differences between groups cannot be primarily attributed to whether patients received the Gemura intervention or not. Length of stay is also likely affected by other, unmeasured variables. Though disease category did not appear to contribute significantly to length of stay, disease severity may have played a role, as found in other studies.³⁸ Patient characteristics at the intervention sites showed much greater diversity in province of origin than the control group, suggesting that patients could have been transferred for higher levels of care several times before admission and had greater disease severity. In addition, the intervention sites had more patients in diagnosis categories of NCDs, cancers, and “not classified,” potentially requiring longer treatments and stays. In addition, since neither clinical recovery endpoint data nor follow-up data after discharge, the results provide a limited perspective on patient recovery. LOS in particular seems to be more affected by many other factors besides clinical and health outcomes and as such, is an insufficient measure of patient recovery on its own. More investigation is needed to elucidate the effect of the Gemura feeding program on patient recovery.

Results on costs of care show that patients spent on average nearly 40% less on food and beverages in the Gemura group compared to the control group, with high statistical significance. This difference in average spend (7,305 RWF or about 5.23 USD) represents a considerable reduction in out-of-pocket costs since Rwanda’s GDP per capita is 966 USD, with 52% of the population living beneath the 2.12 USD per day poverty line.² The higher total costs of stay experienced in the control group is consistent with findings on longer length of stay than the intervention group. But, the average cost savings of 1,948 RWF per day in the Gemura group is still more than half the daily income for the majority of the population. By reducing out-of-pocket costs, these findings indicate the Gemura intervention has the potential to reduce a well-known barrier to healthcare access in Rwanda.

Patient satisfaction survey results showed little difference between groups, were weakly significant, and as a general measure, did not reflect specific perspectives on food provision. Sample sizes, calculated to show significance for the primary outcome, may not have been large enough to show differences between groups.³⁹ Our findings provide little information on patient experiences, which is consistent with recent systematic reviews of using NPS in a healthcare setting as a standalone metric.⁴⁰

Though higher on the single metric visual scale (VAS), HRQOL multi-dimensional scores were only marginally better in the control group for mobility, self-care, and usual activities, and slightly better in the Gemura group for pain/discomfort, providing little information from the patient perspective. The EuroQoL EQ-5D-3L has been used only a few times in Rwanda, with previous studies also finding statistically significant differences in VAS scores but not in the dimensional measures.^41–42 More investigation may be needed on patient experience and perceived quality of life as well as the usefulness of this tool in the Rwandan healthcare context. The most notable difference in our study was in the anxiety/depression dimension, with nearly 40% less in the Gemura group reporting any problems. This indicates that the intervention has the potential to improve patient perception of their mental health during the hospital stay.