Dose of early intervention treatment during children’s first 36 months of life is associated with developmental outcomes: an observational cohort study in three low/low-middle income countries

Study population

Infants with birth asphyxia (resuscitated) and infants without birth asphyxia or other perinatal complications (non-resuscitated), born from January 2007 through June 2008 in rural communities in three sites in India, Pakistan and Zambia, were matched for country and chronological time and randomly selected from those enrolled in the First Breath Trial [34]. Infants were screened for enrollment into the BRAIN-HIT during the 7-day follow-up visit after birth [31], and were ineligible if: (1) birth weight was less than 1500 grams, (2) neurological examination at seven days of age (grade III by Ellis classification) [35], was severely abnormal (because they were not expected to benefit from EDI), (3) mother was less than 15 years old or unable/unwilling to participate, or (4) mother was not planning to stay in the study area for the next three years. Birth asphyxia was defined as the inability to initiate or sustain spontaneous breathing at birth using WHO definition (biochemical evidence of birth asphyxia could not be obtained in these settings) [36]. A list of potential enrollees was distributed to the investigators in each country to obtain written consent for the study, which was obtained during the second week after birth and before randomization to intervention conditions of the BRAIN-HIT.

Intervention procedures

Investigators at each research site selected EDI parent trainers who were trained in an initial 5-day workshop, which was led by the same experts at each research site. A second workshop was conducted before participating children began to reach 18 months of age to adapt the approach to children up to 36 months, again conducted by the same experts at each site. To maintain quality of implementation, the trainers were supervised with observations during actual home visits and constructive feedback was provided on a regular basis.

Each parent–child pair was assigned to the same trainer throughout the trial whenever possible, who was scheduled to make a home visit every two weeks over the 36-month trial period. As elaborated elsewhere [31, 32], the trainer presented one or two playful learning activities during each visit targeting developmentally appropriate milestones. These activities cover a spectrum of abilities across the cognitive, social and self-help, gross and fine motor, and language domains. The parent practiced the activity in the presence of the trainer who provided feedback. Cards depicting the activities were then left with the parent, who was encouraged to apply the activities in daily life with the child until the next home visit. The trainer introduced new activities in subsequent visits to enhance the child’s developmental competencies.

Treatment dose indicators

Two indicators of treatment dose were calculated. Home visit dose was measured based on each parent trainer keeping a record of visit dates. Following the first visit, visits were scheduled to occur every two weeks until the completion of the trial. A home visit was completed on schedule if it occurred within its assigned two week window following the preceding visit. We calculated the percentage of scheduled home visits completed for each participant for the full 36-month trial. The reason for each missed visit was coded as due to illness, weather, death in family, refusal, child or mother unavailable for another reason, parent trainer schedule conflict, and other reasons.

Program implementation dose was measured based on maternal report obtained by the trainer at each home visit of the proportion of days the assigned activities had been implemented since the previous visit. First, the number of days between subsequent completed visits was calculated (Y_n). If the time between two home visits extended beyond 30 days, a maximum of 30 days was used. Program implementation credits were assigned for the time period between visits based on the mother’s report of implementation of activities, as follows: “not at all” (credit_n = 1), “about one-quarter of days or less” (credit_n = Y_n*.25), “about one-half of days” (credit_n = Y_n*.50), “about three-quarters of days” (credit_n = Y_n*.75), and “almost every day or more” (credit_n = Y_n). The credits were then added together over the trial period, divided by the number of possible credits, and multiplied by 100. Thus, this score estimates the percent of days between each home visit that the mother reported implementing child stimulation activities. As an additional descriptive measure of treatment dose, the parent trainer was surveyed at the conclusion of the study to estimate how often the activities had been implemented between the home visits, using a five-point scale (from “never” to “always”).

Developmental outcome measures

The Bayley Scales of Infant Development – II (BSID) [37] was selected as the main outcome measure for this trial because it has been used extensively in various L/LMIC. The BSID underwent pilot-testing at each site to verify validity in the local context and a few items were slightly modified to make it more culturally appropriate (e.g., image of a sandal instead of a shoe). Evaluators across the sites were trained to standards in joint 4-day workshops conducted by experts before each yearly evaluation. The BSID was administered directly to each child by certified study evaluators, who were masked to the children’s birth history and randomization, in the appropriate language with standard material. Both the Mental Developmental Index (MDI) and Psychomotor Developmental Index (PDI) were used to measure developmental outcomes. Scores from the 36-month assessment, obtained just after the completion of the EDI, were used in this analysis as an indicator of treatment outcome.

Health and sociodemographic measures

Perinatal and neonatal health variables were obtained from records kept by the FIRST BREATH Trial [34]: child gender, birth weight (1500 g-2499 g, 2500 g-2999 g, 3000 + g), gestational age (28–36 weeks, 37+ weeks), number of prenatal visits (0, 1–3, 4+), and parity. Additional child health variables obtained as part of this trial at 12 months of age included weight for age/sex (<5th, 5th-14th, 15th + percentile) and complete immunization status.

Family demographic variables were obtained at enrollment in BRAIN-HIT using a structured parent interview: maternal age, education (none and illiterate, none but literate or primary, literate with some secondary), family assets and home living standard. The presence of 11 family assets (e.g., radio, refrigerator, bicycle) were tallied as a Family Resources Index and classified into three levels (0–1, 2–4, 5+). A Home Living Standard Index was calculated based on seven indicators (e.g., home building material, water source, type of toilet) and classified into three levels (0–4, 5–7, 8+). A socio-economic status (SES) measure was used to classify participants into three groups (quintile 1–3, 4, 5) [38].

Statistical analysis

Descriptive statistics were computed for child health and family demographic characteristics, treatment dose indicators (home visits dose and protocol implementation dose), and developmental outcomes (MDI and PDI at 36-months) for all individuals randomized to receive EDI. Child health and demographic characteristics were summarized separately for those randomized to receive EDI and included in the treatment dose analysis and those who were excluded from this analysis, and differences in mean values for continuous variables were tested using t-tests and categorical measures were tested using chi-square and Fisher exact tests. A Pearson correlation statistic was computed between the treatment dose characteristics.

Study sample composition

The sample size was determined to provide adequate power to test EDI treatment efficacy, the primary aim of BRAIN-HIT. As outlined in Figure 1, of 540 births screened from January 2007 through June 2008, 438 (81% of screened) were eligible. Only 3 infants were ineligible due to low birth weight or neurological exam, with the remaining 99 being due to mothers not being able to commit to staying in the study communities or could not be reached for screening within 7 days of birth. Informed consent was obtained for 407 (93% of eligible; 165 resuscitated, 242 not resuscitated) who were randomized into either EDI or a control intervention [20]. The 204 assigned to receive EDI (50.1% of those randomized) are relevant for this study, of whom 145 (71.1% of those assigned to EDI) were included in this analysis (Table 1). These participants had mean = 36.8 (range = 35-41) months of age at the time of the developmental assessment.

Exclusions from this analysis were due to death (n = 7), withdrawal (n = 6), loss to follow up (n = 5), incomplete 36-month BSID-II (n = 39) due to administration errors, home-visit data unavailable (n = 1), or another reason (n = 1). Three children were included in the analysis who completed the 36-month evaluation but discontinued the EDI prior to the end of the study (two because the family had insufficient time to fulfill study requirements and one because the family moved). When compared to those who were included in the analysis (Table 1), children excluded (n = 59) were significantly (p < .05) more likely to have been less than the 5th percentile in weight and completed all immunizations at 12-months of age, and their mothers to have had prenatal care, lower parity, and more family resources.

Description of developmental outcomes and treatment dose

The sample had an unadjusted mean (SD) MDI = 101.2 (10.4) and PDI = 106.8 (14.1) at 36-months. Average home visits dose was 91.4% over 36 months, when 8,990 visits out of 9,841 were completed on schedule every two weeks, and 95% of the participants achieved 80% or greater home visits dose. The most common reason for a missed visit was the inability to locate the mother and child at home at the scheduled time (40.3%), for example because the family was travelling away from the home or had moved temporarily. However, the second most common reason was those related to the parent trainer, such as being ill or having a conflict with another meeting (23.9%). Child or mother unavailable for other reasons (15.3%), for example because the mother was working or baby was sleeping, and weather (10.0%) were the only other reasons accounting for at least 10% of the missed visits. Mother or family directly refusing the home visit at the scheduled time was rare (2.5%).

Mothers reported engaging the child in the assigned activities on an average of 62.5% of days throughout the 36 month period. This protocol implementation dose equates to practicing the intervention activities 4.4 days per week or 674 days over the 36 month trial period. Home visits dose was modestly correlated with protocol implementation dose (r = 0.35). Parent trainers estimated at the end of the trial that 66.2% of families practiced the intervention “always” or “almost always” throughout the 36 months.

Associations between treatment dose and developmental outcomes

Higher home visits dose was associated with higher MDI at 36-months (Figure 2). Specifically, quintiles 1–2 mean MDI = 98, while quintiles 3–5 mean MDI = 103 (Table 2). General linear models of MDI supported this relationship when home visits dose was entered as a primary predictor and site, resuscitation status at birth, and 12-month MDI were entered as covariates (Table 2). Most notably, in the model with only home visits dose (Model 1) and the model which included site (Model 2), mean MDI for quintiles 1 and 2 was significantly lower than quintiles 3–5. A step-down test comparing mean MDI for those with home visit dose below the 40th percentile (quintiles 1 and 2) to those with home visit dose above the 40th percentile (quintiles 3–5), provided estimates of 97.8 and 103.4 (p = 0.0017), respectively . Adjusting by site increased the magnitude of the difference by at least 25% (96.8 vs. 103.9, p = 0.0005). When adjusting for 12-month MDI and the interaction between dose and 12-month MDI (Model 5), the adjusted mean scores for the dose quintiles mirrored unadjusted scores, with quintiles 1–2 consistently lower than quintiles 3–5 (p <0.0001). The lower limit for quintile 3 includes those receiving a minimum of 91% of all the planned home visits.

Based on the same general linear model analysis (Table 2), home visit dose was not significantly associated with PDI at 36 months when considered by itself (Model 1) or when adjusted by site, resuscitation status, and 12-month PDI (Models 2–4). However, there was a positive association between home visits dose and 36-month PDI when adjusting for the 12-month PDI and its interaction with dose (Model 5). Here again, a home visit dose above the 40th percentile (quintiles 3–5) resulted in higher estimated PDI (108.5 – 111.0) compared with below this percentile (103.3 – 106.5)

Higher program implementation dose was associated with slightly higher MDI at 36-months compared to those with a lesser dose. Quintiles 1–2 had a mean MDI of 100 or lower, while quintiles 4–5 has a mean MDI of 102 or higher (Table 2), and the difference appears larger when considering the medians of these quintiles. In a general linear model of 36-month MDI (Table 2), program implementation dose was not a significant predictor by itself (Model 1). However, prediction of program implementation dose when adjusting for 12-month MDI and its interaction with dose (Model 5) indicated that greater dose was associated with higher MDI (adjusted mean Q1 = 100.1 vs. Q5 = 103.1, p = 0.0434). PDI at 36 months was not linearly associated with program implementation dose (Table 2). Rather, mean PDI across quintiles followed a U-shape with the highest mean scores for quintiles 1, 4 and 5. The lower limit for quintile 4 includes those implementing activities on 67% of days on average over the trial period.

Factors associated with treatment dose

The following variables were associated with home visits dose at P ≤ 0.20 when either adjusted by location or by the location by variable interaction: maternal education, parity, family resources, prenatal visits, birth attendant, 1 minute Apgar, preterm birth, and child’s weight at 36-months. These variables were entered into a generalized linear model along with those interaction terms with location that were significant. After backward elimination, the final model (R² = .19) included parity (82.9 ± 3.0 [adjusted mean ± standard error] with 1 child, 79.7 ± 2.8 with 2–3 children, and 90.8 ± 3.5 with 4+ children [p = 0.0382]), 1 minute Apgar (86.9 ± 2.6 for <9 and 82.0 ± 2.6 for 9+ [p = 0.1754]), location (adjusted mean ranged from 75.6 - 94.1, [p = 0.0019]), preterm [(p = 0.4571) and preterm by location interaction(p = 0.0020). There was a substantial difference in relationship to home visits dose by prematurity across location. Location A had higher dose for term children (65.8 ± 6.3 for preterm and 85.3 ± 4.0 for term). Location B had essentially the same dose between groups (92.8 ± 5.9 for preterm and 95.4 ± 2.9 for term). Location C had considerably higher dose in preterm children (90.5 ± 4.6 for preterm and 76.9 ± 3.5 for term).

The following variables were associated with program implementation dose at P ≤ 0.20 when either adjusted by location or by the location by variable interaction: home visit adherence rate, maternal education, parity, family resources, living standard index, prenatal care, 1 minute Apgar, preterm birth, and weight at birth, 12, 24, and 36 months. These variables were entered into a model along with those interaction terms with location that were significant. After backward elimination and adjusting for location, the final model (R² = .25) included home visit adherence rate (a one percent increase in home visit adherence resulted in a 0.64 ± 0.18 percent increase in program implementation adherence, p = 0.0004), maternal education (70.0 ± 2.8 for secondary/university and 60.9 ± 2.4 for none/illiterate [p = 0.0400]), prenatal care (71.0 ± 2.9 for 5+ visits and 65.3 ± 3.5 for no care [p = 0.0170]), weight at 12 months (66.7 ± 1.7 for >85th percentile and 61.1 ± 2.2 for <5th percentile [p = 0.0917]), and location (adjusted mean ranged from 59.5 - 69.1, [p = 0.0019]). None of the interaction terms were retained in the final model.