1. Introduction

Two decades ago, in 2004, the World Health Organization (WHO) and UNICEF first called for childhood diarrhoea to be treated with both oral rehydration salts (ORS) and zinc sulphate [1], estimating, at that time, that 1.5 million children under five years of age were dying of diarrhoea annually. Co-therapy with zinc sulphate was recommended following studies that found that its administration reduced the severity and duration of diarrhoea in addition to offering a protective effect from reinfection [2,3]. ORS has been recommended to prevent death through rehydration since the 1970s [4]. While mortality rates due to diarrhoea have improved over the past 20 years, coverage of diarrhoea cases with both ORS and zinc still stands at just 18% in sub-Saharan Africa and 19% globally [5] despite some efforts to improve it, particularly through supply chain stimulation coupled with awareness raising and policy guidance [6–17]. Diarrhoea continues to cause an estimated 444,000 deaths annually, equating to 9% of all deaths among children under the age of 5 years worldwide [5].

In 2019, WHO accepted an application from key actors [18] and amended its Model List of Essential Medicines to include, for the first time, co-packaged ORS and zinc (ORSZ) for the treatment of diarrhoea [19]. Key points in the application included the potential of co-packaging to increase coverage and improve adherence to the combined therapy as, particularly in low-resource settings, sachets of ORS and zinc sulphate tablets may not be available at the same time and in the right ratios, or health personnel may not know of the recommendation to dispense both medicines together. Co-packaging was accepted by WHO as a valid approach to increase coverage of the co-therapy and is gaining ground as a possible intervention, adaptable to local contexts, to address this implementation gap [8,12]. UNICEF states that over 20 countries have introduced co-packaged ORSZ in the public sector and now advocates that countries use only co-packaged ORS and zinc to promote adherence and optimal treatment [12]. However, only a few studies have been conducted to explore or validate the effect of co-packaging ORS and zinc on the coverage of ORSZ combined therapy [9–11,13–16,20]. Black (2019) concluded that co-packaging can rapidly equalise the use of ORS and zinc [8], and some governments are reportedly moving towards a preference to procure co-packs [9–11,14]. Few studies, however, have specifically explored the effect of co-packaging on dispensing practices.

Here, we investigate whether the introduction of a co-pack containing ORSZ at remote rural government health facilities in Zambia affected the dispensing practices of health personnel. We compare dispensing practices at various health facilities before and after the introduction of a new co-pack by the Zambian government. All facilities were supported by the District Health Office in Mongu town (Mongu District). The intervention formed part of a larger trial and subsequent scale-up initiatives of co-packaged ORSZ, which were conducted in Zambia from 2011 to 2018 and are reported elsewhere [14–16,21]. The aim of our analysis was to assess whether the introduction of the co-pack significantly increased the likelihood of a child under five years of age, who is affected by diarrhoea, being prescribed and dispensed both ORS and zinc, on presenting with diarrhoea at a facility.

2. Methods

2.4. Statistical methods

For the purposes of this study and based on the global WHO/UNICEF recommendation, we considered the dispensing of diarrhoea treatment as correct when both ORS and zinc were provided to the patient or their caregiver, either packaged individually (the only option in 2016) or as a co-pack. Accordingly, we clustered dispensing behaviour into a dichotomous response, distinguishing between ‘correctly-dispensed’ and ‘incorrectly-dispensed’ cases. The aim of the present work was to investigate whether the correct dispensing rate (CDR) underwent a change, following the introduction of the co-pack.

The data presented a two-level structure, with individual dispensing behaviour nested within facilities. We analysed the data using multilevel logistic modelling, to account for potential correlation in dispensing practices within the same facility [27] and to draw reliable inference from health facilities with fewer cases. Specifically, we used two models, to achieve two different goals: a simple random-effect model to estimate the CDR of each facility in either of the two years; a more complex mixed-effect model to quantify the impact of the co-pack introduction on the facilities’ CDR and odds ratio. Details on the models and the rationale behind the choices follow. All models were fitted in R, version 4.3.2, via the lme4 package.

All models trained in this study were logistic regression models with dichotomous response identifying correct/incorrect dispensing behaviour. Thus, each model predicted the odds of correct dispensing (the ratio between the probability of correct dispensing and the probability of incorrect dispensing) as a function of relevant predictors. The second of the following two formulas (one the inverse of the other) then allowed to expressed the odds of correct dispensing as CDR:

(1)

In the random-effect logistic model, facilities were included as random effect and no fixed effect predictors were included. We fitted the model, separately, on each of the two year’s data. This enabled us to estimate the odds (and thus the CDR) of the single facilities, and the associated uncertainty, in each of the two years separately (2016, no co-pack available; 2017, co-pack available). We notice that, through the random-intercept approach, each facility’s CDR was inferred by adjusting the overall CDR to the data observed in that facility, thus compensating for the limited amount of data of some individual facilities.

After estimating the CDR of the single facilities, we fitted a mixed-effect logistic regression to the whole data set. The model included a factor for co-pack availability as fixed effect and one for facilities as random effect. In other words, a linear predictor was used to model the log odds of correct dispensing, with fixed slope quantifying the effect of co-pack availability, and random intercept varying among facilities. The choice allowed us to explicitly model the effect of the co-pack introduction on dispensing practices, while accounting for random variation between facilities due to factors which could not be measured.

For both models, we quantified the average degree of correlation among dispensing practices within the same facility via the intraclass correlation coefficient (ICC). We computed the ICC using the simulation approach described in [28], as appropriate to the case of binary-outcome logistic models. A near-zero ICC indicates that a lack of within-facility correlation may be assumed for modelling purposes, and thus that a classical logistic model (with no random effects) may also be used. We will consider this option during the exposition of the results, according to the outcome of the multi-level model.

3. Results

For each of the seven facilities that met our inclusion criteria, Table 1 reports the number of cases for which a diarrhoea treatment was dispensed, and the proportion of these that were correctly dispensed (i.e., that included both ORS and zinc). The data is shown for each of the two years separately (2016 and 2017, before and after the co-pack introduction, respectively), with the last row summarising results across all facilities.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Observed diarrhoea cases and correctly dispensed cases (in absolute and relative terms) in October 2016 and October 2017, by facility and overall.

https://doi.org/10.1371/journal.pgph.0004342.t001

The overall CDR in the sample, as computed across all facilities, increased from 44% to 87% following the introduction of the co-pack. As Table 1 shows, the CDRs of individual facilities vary around the aforementioned percentages. We estimated the underlying CDR of each facility, separately in each year, via a random-intercept model with correct dispensing as binary outcome and facilities as random intercept. Table 2 presents the results of fitting the model to the 2016 data (before co-pack availability) and to the 2017 data (after co-pack availability). Fig 1 offers a graphical illustration of the change in the estimated CDRs (bars) and associated CIs (whiskers), between the two years.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Results of the random-intercept model with correct dispensing as binary outcome and facilities as random intercept.

https://doi.org/10.1371/journal.pgph.0004342.t002

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Comparison of the CDR at each of the seven health facilities, before and after the introduction of co-packaged ORSZ.

The CDRs (bars) and associated CIs (whiskers) are estimated through random-intercept models with correct dispening as outcome and facilities as random intercept (Table 2).

https://doi.org/10.1371/journal.pgph.0004342.g001

The bar plot in Fig 1 reveals that all facilities increased their CDR following the introduction of the co-pack. The varying amplitude of the CIs reflects different sample sizes at each facility in each year (narrower CIs generally correspond to larger sample sizes). Notice that, with the exception of Nalwei RHC, the CIs for the two years at any facility do not overlap with each other, indicating that the increase is significant at least at the 5% level. At Nalwei RHC, a small overlap can be detected. Additionally, we note that for Mulambwa UHC and Nalwei RHC the increase is present, but more contained than for the remaining five facilities. In contrast to the other facilities, these two had relatively high CDRs prior to the co-pack introduction.

As indicated by Table 2, the simulated ICC for the models is equal to 0.161 in 2016 and 0.077 in 2017. This reveals that around 16% and 8% of the total variability in dispensing practices in the two years can be attributed to differences between the facilities, while most of the variability comes from varying dispensing practices within facilities.

To quantify the change in CDR that took place following the introduction of the co-pack, and its statistical significance, we fitted a mixed-effect model to our data. The model has co-pack availability as fixed effect and facilities as random intercepts.

Table 3 summarises the results of the fit. The common odds ratio across facilities is estimated to be 8.42 (95% CI: 5.47 - 12.9). The results suggest that the odds of correct dispensing at least quintupled following the introduction of the co-pack. The odds ratio estimate is highly significant, as is the whole model: a likelihood ratio test of the model versus an intercept-only model (both with fixed intercept, and with facility-dependent random intercept) returns a p-value<0.0001.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Mixed-effect model results (co-pack availability as fixed effect, facilities as random intercepts).

https://doi.org/10.1371/journal.pgph.0004342.t003

It is interesting to note that we estimate a near-zero ICC () for the model. This reveals that, once a factor for co-pack availability is taken into account, the data behaves as though dispensed cases within facilities were independent. This would make a standard logistic regression, with co-pack availability and facility as covariates, also appropriate to the dataset. In additional investigations we do not detail here, we have also explored this option: as expected, the results confirm the outcomes of the above mixed-effect model, hardly adding any relevant information to it (highly significant co-pack factor, not significant facility in the presence of co-pack).

From Fig 1, we note, however, that two groups of facilities may be identified, according to the CDR that these displayed prior to the co-pack introduction: Mulambwa UHC and Nalwei RHC on the one hand (high CDR prior to the introduction of the co-pack, around 60%) and the remaining five facilities on the other (low CDR prior to the introduction of the co-pack, below 30%). We thus decided to fit a classical logistic regression model to the data of each of the two groups separately. The model included co-pack availability and facility as covariates. This aimed to investigate whether the estimated odds ratios in the two cases would be different from each other, while no significant difference would be detected between facilities in the same group.

The results confirm the above intuitions. The odds ratio is estimated to be 2.17 (CI: 1.24 - 3.77, p=0.006) for the high-baseline-rate group, and 128 (CI: 44 - 493, p<0.0001) for the low-baseline-rate group. Note the much wider CI in the second case, due to the lower amount of data and a much greater central estimate of the odds ratio. While the odds ratio estimates are very different for the two groups, no significant difference shows up between facilities of the same group. The p-value for the difference in log odds between Mulambwa UHC and Nalwei RHC is 0.96. Similarly, all p-values for the difference in log odds between pairs of facilities of the second group are higher than 0.19. We speculate on the possible causes of this in the discussion.

4. Discussion

The transition from individual ORS and zinc products to a co-packaged format, particularly within the public sector, has emerged as a pivotal strategic initiative across several countries, significantly elevating combined ORS and zinc coverage levels [8–11,13,14]. This is one of the first studies to look specifically at the influence of co-pack introduction on dispensing practices at the facility level.

Overall, across the seven frontline health facilities analysed in this study, the introduction of co-packaged ORSZ increased the rate of dispensing of the recommended treatment for diarrhoea (ORS together with zinc) from 44% to 87%, corresponding to an 8.42 increase in odds ratio (95% CI: 5.47 - 12.9, p<0.0001). Disaggregation of the overall improvement showed that all facilities in the study improved their dispensing of the recommended treatment for diarrhoea following the introduction of co-packaged ORSZ. The largest increases in correct dispensing rate (CDR) were seen in the less well-resourced HPs (Kaande, Mweeke, Nasange), as well as in two of the three RHCs (Lukalanya, Lukweta). Despite a lower number of cases across the five facilities (86 in 2016, 122 in 2017), the estimated common odds ratio for the facilities was highly significant (128, CI: 44 - 493, p<0.0001). The increase in odds ratio was present and statistically significant also for the other two facilities (Mulambwa UHC and Nalwei RHC), although more contained (2.17, CI: 1.24 - 3.77, p=0.006). This suggests that co-packaging ORSZ may be particularly helpful in less well-resourced settings, where fewer and/or less-trained staff may be available.

The retrospective nature of this study meant that qualitative interviews with staff were not possible. Key informant interviews would have been helpful to understand certain aspects of the dispensing behaviour observed at Mulambwa UHC and Nalwei RHC. These two facilities displayed the highest CDRs before the introduction of the co-packs and benefited the least from their introduction. They also dispensed, in 2017, a relatively low percentage of co-packs out of all correct treatments: 88% and 69% respectively, compared to the 100% at four of the other five facilities (Table 1). Various factors could explain this. It may have been that the staff at these facilities were better trained [29,30] and that dispensing was determined more closely by expiry dates such that older stocks of ORS and Zinc, packed separately, were dispensed prior to the newly arrived co-packs. It may have been that more experienced or better trained staff had less need to embrace the convenience of the co-pack. However, our data does not include the qualitative information required to draw robust inference on potential reasons behind this observation.

We note that, while improved dispensing can be expected to translate into improved coverage, other factors such as user preferences [31], rational use, design and adherence [15] are known to influence effective coverage [8,32,33]. Nevertheless, the CDR of 87% seen in this study on a sample of about 400 cases compares very favourably to the Zambian national average coverage of ORS and zinc (34%) [24].

We conclude by acknowledging some limitations of the present study. Mongu District was not selected to be representative of all rural districts in Zambia, the choice was mainly dictated by contingent situations pertaining to the wider project discussed at the end of the Introduction section. The restricted geographical focus limits the extension of the validity of the described results to larger, national or international, geographical scales. Furthermore, the retrospective design of the study precluded sample size estimation prior to data collection, limiting our ability to correct for imbalances between facilities and ensure adequate power to detect predefined effect sizes. Despite these limitations, the results presented here offer robust insights into the research question, achieving high statistical significance with the available data. We believe the evidence is sufficiently promising to warrant further research to externally validate these findings in different contexts and at broader geographical scales. Future studies could also incorporate interviews with dispensing staff to better understand additional factors influencing dispensing behaviour.

5. Conclusions

Our analyses show that, across the examined facilities, the introduction of co-packaged ORS and zinc was highly effective in increasing the number of cases where the globally recommended treatment for diarrhoea was dispensed. The most significant improvements were observed in less well-resourced settings. We encourage continued research to assess the impacts of co-packaging and to further validate these findings on a broader scale.

Supporting information

Acknowledgments

The authors acknowledge the key role of implementation partners not mentioned as authors: Ezra Banda and John Msimuko, the current and former Chief Executive Officers at Keepers Zambia Foundation, respectively; and the field implementation team at Keepers Zambia Foundation, in particular, Ruth Mitimingi, Albert Saka and Chibale Phiri Malekani. We gratefully acknowledge insights and expertise provided freely over the 8 years of our work by members of the wider project’s learning and steering group (LSG), including representatives of various departments of the Ministry of Health at the national and district levels. We thank the chairs of the LSG, Dr Elizabeth Chizema, Dr Max Bweupe and Dr Kennedy Malama, and acknowledge the faith placed in our work by the late Minister of Health, Dr Joseph Kasonde. We thank the Statisticians for Society initiative of the Royal Statistical Society, through which the statistical collaboration on this paper between DD and the other authors was initiated. DD warmly thanks Jacqueline Birks for her support and the fruitful exchanges which have informed the current work. Finally, the authors would like to thank the reviewers, Dr. Charles P. Larson and Dr. Rukhsana Haider, who, through their helpful comments, have contributed to improve the quality of the present manuscript. Editorial support for the original draft, according to Good Publication Practice guidelines, was provided by Adam Bodley.