The trial was conducted in 82 rural Andean communities (registered populated centres) from the San Marcos and Cajabamba provinces, Cajamarca region, northern Peru. Both sites were high-altitude resource-limited locations. The majority of the population were small-scale farmers living in households with adobe walls, and using traditional biomass stoves or open fires for cooking. A detailed description of the study setting is found in Hartinger et al. [13].

Study design

We implemented a cluster-randomised controlled trial to evaluate two home-based interventions: (i) a home-environmental package comprising a kitchen sink, hygiene education, and a certified improved biomass cookstove (ICS) (henceforth referred to as “IHIP”); and (ii) an ECD programme (henceforth referred to as “ECD”). The design led to four potential experimental conditions: (i) IHIP & ECD (henceforth referred to as “IHIP + ECD”), (ii) IHIP, (iii) ECD, and (iv) Control. A detailed description of the study design is found in Hartinger et al. [13].

The interventions comprising the IHIP were selected based on the results of a previous trial in the region [14, 15]. We selected the ICS model after a comprehensive community consultation [16]. Stoves were built with local materials and sinks were purchased locally. Participants received monthly visits during follow-up to reinforce hygiene education and the correct maintenance of ICS. The hygiene education component conveyed three main messages: (i) keeping kitchen environments clean; (ii) washing of mother and child’s hands with soap at key moments (e.g., before preparing meals); and (iii) household water treatment. We promoted boiling, since it was the method endorsed by local health authorities.

For the ECD intervention, we adapted the home-visiting component of the Peruvian National ECD programme (“Programa Nacional Cuna Más”—PNCM). Women living in the participating communities (mother facilitators—MFs) were trained to conduct weekly play-oriented, semi-structured activities with study children in the presence of their caretakers [17].

The study included families that (i) had at least one child < 1.5 years living at the household; (ii) used solid fuels as their main energy source for cooking/heating; (iii) had access to piped water in the yard; iv) did not plan to move within the next 24 months; and (iv) did not participate (but met inclusion criteria to do so) in the PNCM.

Sample size calculation

We assumed three episodes of diarrhoea per child-year in the control arm, and a 25% reduction in incidence in the treatment arm. With 10 person-years of follow-up in each cluster and a coefficient of variation of 0.2, we estimated that 16 clusters for the intervention and control arm were sufficient to detect the anticipated reduction of incidence with a power of 80% at the 5% two-sided significance level. For the ECD intervention, we used the ECD outcome (percentage of tasks solved above the mean of the study population) of our previous intervention study [18] and assumed 60% above mean for the intervention and 40% above mean in the control arm. Using the equivalent formula for proportions, we calculated that 15 clusters for intervention and control were sufficient to detect the differences in ECD status with a power of 80% at the 5% two-sided significance level. To account for potential loss to follow-up, we included a total of 40 clusters (10 clusters per arm) in the study. The trial was sufficiently powered to compare each intervention against its control arm but it did not allow pairwise comparisons among the four arms. A detailed description of the sample size calculation, which followed the formula proposed by Hayes and Bennett [19], is found in Hartinger et al. [13].

Randomisation and masking

Details on randomisation and masking are provided in Hartinger et al. [13]. In brief, the enrolled communities were aggregated into community-clusters based on their proximity to each other. We used a covariate-based constrained randomisation when allocating the communities into the four study arms [20]. First, clusters were divided into 8 strata of 4 clusters and 1 stratum of 8 clusters. Then, we generated two million random allocation sequences and selected those for which the maximum difference between arms fulfilled certain criteria (e.g., number of children, median community size or access to electricity within the community). Of the 164 allocation sequences that fulfilled all criteria, one was randomly selected. Given the trial design and nature of the interventions, the study was not blinded. Contamination of control communities was mitigated given intervention and control communities were geographically separated. Also, we conducted monthly household visits to identify possible structural changes in control households (i.e., installation of sinks). The ECD intervention required home visits and the use of age-adapted toys; contamination of control homes were considered unlikely to occur.


We carried out a census in 2015 in collaboration with local health authorities to identify potential communities, children and pregnant women in their second and third trimester. Participants were enrolled between September 2015 and January 2016. From the screening census, we identified 102 communities with 574 potential children. During enrolment, 237 families were no longer eligible. We re-enrolled participants between January and February 2016 because 21 families were not available or declined to participate in the project at the beginning of the follow-up. A group of seven trained fieldworkers supervised by the field coordinator team enrolled participants. In total, 317 households from 82 communities participated in the trial (Fig. 1). A detailed description of the enrolment procedure is found in Hartinger et al. [13].

Fig. 1

Flowchart of the cluster-randomised controlled trial. aTwo children without any follow-up information excluded from the final analysis. bFour children without any follow-up information excluded from the final analysis. cOne child without a matched counterpart in its age group (18 months) was excluded from the final analysis. dIt refers to diarrhoea morbidity


The IHIP-2 trial was conducted between April 2016 and May 2017. We visited all households weekly and collected daily and weekly self-reported information from the mother or caretaker about the occurrence of signs and symptoms of diarrhoea and ARI. During household visits, we measured respiratory rate, heart rate, and oxygen saturation in blood (SpO2) with portable pulse oximeters. Severely ill children were referred to local healthcare facilities for further evaluation. We collected health and anthropometric data monthly from participant’s clinical records at local health centres. When assessing ECD status at end-of-study, we applied two instruments i) the nationally validated Peruvian Infant Development Scale (ESDI) [21] and the Spanish version of the internationally validated Bayley Scales of Infant and Toddler Development (BSID) [22]. The ESDI tool was designed by the PNCM based on developmental studies conducted by the World Health Organization (WHO) and the recommendations of an expert panel. It assesses ECD status of children aged 1 to 36 months through direct observation, interaction or caregiver’s self-reports [21]. At end-of-study, some children were over the maximum age of 36 months at which age the ESDI tool can be carried out. The main reason was that the start of the trial needed to be postponed due to the re-enrolment. The BSID tool evaluates ECD status of children aged 1 to 42 months using a series of play-tasks [22]. The application of the ESDI allows the results to be used by the PNCM in its effort to expand the programme nationwide. We applied the BSID tool for comparability reasons despite differences between tools in age range, as the BSID is the current gold standard for measuring ECD status. A group of four fieldworkers conducted the ESDI assessments. They received a one week training from PNCM experts and were supervised by the field coordinator team. The BSID tool was applied by another group of four psychologist from the Universidad Peruana Cayetano Heredia (UPCH). Their work was supervised by a head psychologist who had previous experience applying the BSID. Information was revised on a daily basis to reduce the chance of missing data. We carried out weekly spot-check observations and obtained maternal self-reports to assess compliance with the IHIP, and monthly maternal self-reports for the ECD intervention. We collected 24 h stationary air pollution data in the kitchen using carbon monoxide (CO) monitors (EL-USB-CO, LASCAR Electronics, Whiteparish, England) and fine particulate matter (PM2.5) devices (APROVECHO-5000, APROVECHO Research Center, Cottage Grove, USA) at a one-meter distance from the ICS and at standard breathing height (1.5 m). Household air pollution (HAP) data were obtained from a sub-sample of 40 participants (henceforth referred to as “sentinel sub-sample”) on five occasions (before ICS installation, three times during follow-up, and at end-of-study). Water samples were obtained from the child’s main drinking source. They were collected at baseline and end-of-study for all study participants and in the sentinel sub-sample on three additional occasions during follow-up. We used a membrane filtration method for identifying thermo-tolerant faecal bacteria (DelAgua Water Testing Ltd, Marlborough, UK). All yellow colony-forming units were considered positive for bacteria growth, and microbial contamination was determined applying the WHO standards of zero viable coliforms. Finally, we administered a socio-economic questionnaire at baseline and end-of-study to assess household demographics, education and economic characteristics. We used the nationally validated Young Lives Wealth Index to classify participant’s wealth status [23]. A detailed description of the study’s methodology, field operations and procedures is found in Hartinger et al. [13].


Diarrhoea and ECD status were the primary outcomes. We defined diarrhoea following the WHO definition of the passing of at least three loose stools within 24 h. We considered an episode to begin on the first day of diarrhoea and to end on the last day of passing a diarrhoeal stool, followed by at least three consecutive diarrhoea-free days. We defined ECD outcomes as the age standardised mean scores of socio-emotional, fine and gross motor, communication, and cognitive skills and an overall performance, defined as the arithmetic mean of the five categories. Secondary outcomes included: (i) ARI, defined (according to WHO standards) as presence of cough and fever. We defined an ARI episode to begin on the first day with cough and fever, ending on the last day with symptoms followed by at least seven symptom-free days; (ii) severe cases of diarrhoea, defined as persistent diarrhoea (14 days) or bloody diarrhoea; (iii) kitchen levels of CO and PM2.5 in the sentinel sub-sample; (iv) presence of thermo-tolerant faecal bacteria in drinking water samples; (v) stunting and underweight, defined according to WHO standards; and (vi) compliance linked to the interventions. We defined stove compliance as keeping the ICS structure and chimney in good condition (i.e., without deep cracks and not dissembled) and observing ICS use at the time of the visit or reporting ICS use in the last 24 h. Sink compliance was defined as keeping the structure in good condition and observing the presence of soap or dishes on the sink at the time of the visit. We defined ECD compliance as a reported ECD session since the last supervision visit.

We specifically selected diarrhoea and ECD status as primary outcome due to the important effort made in the international community to explore the effects of hygiene and ECD interventions combined [8, 12]. On the other hand, we selected ARI as a secondary outcome due to the low incidence of respiratory diseases detected in the area in previous studies [10], and the available evidence on the effect of ICS to detect improvements in children’s health [24].

Safety of the interventions was assured; the ICS was certified by the Peruvian National Training Service for the Construction Industry (certificate number: 04-2015-LCM-GIN-SENCICO), and the ECD intervention followed the PNCM protocol. Parents were instructed to consult the nearest health centre for any health concern. Health centre contacts were assessed at each round of household visits and recorded if they were related to the trial intervention.

Statistical analysis

Data were entered in the CSPro 6.1 database (U.S. Census Bureau, ICF International, Serpro S.A.) and cleaned, prepared and analysed the data using STATA 15.0 (Stata Corporation, College Station, TX, USA) and R 3.4 (R Foundation for Statistical Computing). For diarrhoea and ARI outcomes, we compared number of episodes and illness days. Because the scores of the ECD outcomes might be age dependent, we first calculated the mean score separately for each age category and dichotomised the outcome as performance above or below the age specific mean. For water samples, we compared the total number of samples with positive thermo-tolerant faecal bacteria.

To account for potential correlation within clusters, we employed generalised estimating equation (GEE) models with robust standard errors and an independent correlation structure. The correlation structure was pre-specified and not data driven. It has been suggested that independent correlation structures provide more robust results (compared to the exchangeable) if the number of participants varies among clusters, as was the case in our study [25]. For binary outcomes, we used the binomial family with logit link. For count data (number of episodes), we used a negative binomial distribution with log link and the natural logarithm of the number of days under observation as offset variable. The unadjusted model included the design factors and intervention effect. Further models were adjusted for child’s age and sex. All models were pre-specified in the trial protocol. Other baseline characteristics were of demographic nature. They were included in the covariate-based constrained randomisation but not in the adjusted statistical model. The primary analysis was performed according to the intention-to-treat principle using the full analysis set, i.e., all randomised and re-enrolled children with at least one day of follow-up information. We used the available case population for the analysis. No imputation of missing data was performed due to the low proportion of missing data for diarrhoea related outcomes (only 2.5% of children with no follow-up data) and because the proportion of missing data was balanced among trial arms.

By admin

Slot Gacor Terbaru

Link Slot Gacor


Situs Slot Gacor

Slot Gacor 2022

Login Sbobet

Daftar Sbobet

situs slot gacor

slot gacor gampang menang

slot gacor terbaru

daftar sbobet

slot bonus newmember

link sbobet

sbobet login