A Text-Based Intervention to Promote Literacy: An RCT

We screened electronic charts of patients aged 3–4 years with appointments between October 2017 and August 2019 in 2 large, county-based clinics. Children were ineligible if they had developmental delay, if the family language was not English or Spanish, if there were more patients than the research assistant (RA) could get to, or if the child was deemed too ill. Eligible families were approached for enrollment, and those who consented completed a demographic survey assessing child age and birth order, household structure (single or 2-parent), primary caretaker’s age, role (mother, father), highest education level, race and ethnicity, preferred language, household income, preschool enrollment, and child’s insurance type. Children in preschool or transitional kindergarten were then excluded. Remaining families were randomly assigned through block randomization stratified by parent language and child baseline age.

We measured early child literacy with the Phonological Awareness Literacy Screening Tool (PALS-PreK), which the University of Virginia developed and validated for a diverse child population.⁴⁴  The PALS-PreK measures preschoolers’ knowledge of literacy fundamentals: name writing, alphabet knowledge, sound awareness, print and word awareness, rhyme awareness, and nursery rhyme awareness (Supplemental Table 5). PALS-PreK has revealed reliability, with strong internal consistency measured by Cronbach’s α (0.75–0.93) and similar Guttman split-half scores (0.71–0.94).⁴⁴ 

We calculated composite PALS-PreK scores for Spanish and English separately by summing individual items. Because the 2 PALS-PreK versions slightly differ in content, we standardized composite scores for pre- and postscores by language. Additionally, pre- and postintervention scores were standardized, allowing for comparison, creating the possibility of negative preintervention scores. As in previous research,⁴⁵  we did not include letter sounds and lowercase letters questions in the composite score because children must score well on previous sections to be given these items, and most did not qualify.

Parent–child language exchanges were measured using the Parent Child Interactivity Scale (PCI), which uses parent self-assessment and has been validated in similar populations.⁴³  The PCI uses a 4-point Likert scale for 5 items assessing reading, such as “Looked at pictures together in a book,” and 10 items assessing other literacy activities, such as “Practiced rhyming words.” (Supplemental Table 6) The PCI has revealed internal consistency in English (Cronbach’s α = 0.83) and Spanish (Cronbach’s α = 0.84).⁴⁵  We calculated reading (PCI Read), other activities (PCI Activities), and a composite score (PCI Total).

The research team conducted weekly check-ins to monitor data collection. Improvement procedures included standardized protocols and monitoring of interrater reliability over 9 months (May 2018 to February 2019).⁴⁶  We conducted interrater reliability for 17 PALS-PreK assessments (5% of assessments) across 4 independent raters (1 primary, 3 secondary). The interclass correlation coefficient ranged from 0.97 to 1.0 across raters with no change over time. Bilingual RAs administered the PALS Pre-K, and the vast majority of RA assessments were blinded. Unblinding occurred when families mentioned receiving texts or when an RA had to cover a different team role; sensitivity analyses indicated no impact on results.

Intervention caregivers received texts supporting literacy promotion 3 times a week for 7 months.⁴³  The texts were designed by educational professionals as a 2-generation, caregiver empowerment model drawing on family strengths. Texts were structured to build on one another and delivered in small bits, delivering a light cognitive load to the caregiver. Each week, adults received 3 texts about an academic skill or set of skills: a “FACT” text designed to inform and motivate caregivers; a “TIP” text to minimize the cognitive, emotional, and time burdens of high-quality parenting by providing caregivers with specific activities building on existing family routines; and a “GROWTH” text to provide encouragement and reinforcement. (Supplemental Table 7) Each text began with “Doc Says” to convey the pediatrician’s support. The control group did not receive texts.

We used descriptive statistics to summarize baseline child, caregiver, and household characteristics. We implemented univariate analyses using χ² tests and Fisher’s exact tests for categorical data and t tests for continuous data to determine if excluded and included participants differed. We similarly assessed differences between control and intervention groups to test whether randomization was effectively matched.

For PALS-PreK and PCI, we calculated summary statistics at baseline (preintervention) and at follow-up 7 months later (postintervention). We used multivariable regression to compare postintervention scores between control and intervention participants and calculated 95% confidence intervals for estimated effect sizes. For each outcome, we ran 3 models: model 1 was used to estimate the bivariate relationship between treatment and outcome (unadjusted), model 2 was adjusted for baseline score, and model 3 was adjusted for baseline score, age in months, and caregiver language. We further conducted subgroup analyses to assess the heterogeneity of effects, examining child birth order (firstborn, not firstborn), family structure (single-parent family, not single-parent family), caregiver language (English, Spanish), caregiver education (less than high school, high school, some college or more), and child baseline age (≥48 months, <48 months). Because of the skewed distribution of the standardized PALS-PreK score, asymptotic normality may not hold, given finite samples.^47,48  Therefore, we calculated the standard errors of our model estimates using bootstrapping for valid inference.⁴⁹  Statistical analysis were conducted by using R version 3.6.2 (R Foundation, Vienna, Austria).⁵⁰  All statistical tests were 2-sided and performed at .05 significance level.

The Stanford University Human Subjects Review Board and the Santa Clara Valley Medical Center Research and Human Subjects Committee approved the study.

We screened medical records for 1968 upcoming clinic appointments, identifying 914 potential participants (Fig 1). The majority of ineligible appointments were excluded because of developmental delay (n = 554, 53%). Of the 914 eligible participants, 270 declined, most for the reason of “not enough time on appointment date” (n = 153, 57%). Of eligible participants, 644 completed the demographic survey, after which we excluded an additional 263 (41%) because of preschool or transitional kindergarten enrollment. We randomly assigned 381 participants to the intervention or control group, and 10 caregivers opted out of receiving texts. At the postintervention time, 275 (74%) completed the postintervention assessment, 29 (8%) declined to participate, and 67 (18%) could not be reached. Those who did and did not participate at postintervention (n = 106 lost to follow-up) were similar in terms of sibling order, total children, caregiver age and race and ethnicity, household income, and insurance type (Supplemental Table 8). The groups differed in some respects: those lost to follow-up had slightly older children (43 vs 45 months) with more caregivers who were fathers, spoke English, and attended college. We found no evidence of difference in attrition between intervention and control groups in the proportion leaving or in the characteristics of those who left.

Included participants and those ineligible largely because of preschool enrollment were similar in family structure and caregiver age but differed across several sociodemographic characteristics (Table 1). Eligible participants came from families with lower income and lower educational attainment (65% vs 83% with a high school education or greater). They were slightly younger and more likely to have a Spanish-speaking caregiver (46% vs 29%).

The intervention and control groups had similar demographic backgrounds, family composition, and socioeconomic status (Table 1). Children were an average of 44 months old, and 89% attended clinic with their mothers (average age 33 years). Participants were predominantly Hispanic (73%) with a minority identifying as non-Hispanic white (6%), Black or African American (5%), Asian American (7%), multiracial or ethnic (6%), or other (4%). Almost one-half of participants indicated Spanish was their preferred language. Participants came largely from low-income, 2-parent families (64% with a family income <$36 001; 71% from a 2-parent family) with Medicaid insurance (88.7%). Baseline standardized PALS-PreK scores were similar for control and intervention groups on average (−0.38 [SD = 0.76] vs −0.32 [SD = 0.83]), as were total PCI scores (24.40 [SD = 9.68] vs 24.08 [SD = 9.97]). Postintervention, standardized PALS-PreK scores were, on average, 0.36 (SD = 0.98) for the control group compared with 0.62 (SD = 1.08) for the intervention group. Total PCI scores were 25.96 (SD = 10.0) for the control, versus 28.34 (SD = 9.15) for the intervention group.

TbT’s impact on the children’s literacy development is revealed in the intervention’s estimated effect on follow-up PALS Pre-K scores (Table 2). Model 1 gives the unadjusted overall intervention effect of 0.260 (95% CI: 0.121 to 0.507, P = .040). Model 2 is adjusted for the baseline PALS-PreK score, with estimated effect of 0.202 (CI: 0.031 to 0.372, P = .021). Model 3 is adjusted for baseline PALS-PreK score, baseline child age, and caregiver language, with an estimated effect size of 0.209 (CI: 0.038 to 0.378, P = .016). Including the additional controls had little effect on the point estimates but reduced the standard errors and, as a result, made the estimates more precise. In Model 3, although caregiver language being Spanish was not associated with postintervention PALS-PreK score (−0.053, CI: −0.219 to −0.113, P = .532), child age did have an independent contribution (0.029, CI: 0.014 to 0.043, P < .001), as did baseline PALS-PreK score (0.813, CI: 0.702 to 0.923, P < .001).

To provide real-world context for TbT effects, we compared them to typical learning rates. When we estimate the relationship between child age and postintervention PALS-PreK scores for those in the control group, the estimate is 0.066 (CI: 0.043 to 0.089, P < .001). Thus, in a normal routine, children in our sample would have gained 0.066 points on the PALS-PreK assessment per month. Those in the intervention group gained an additional 0.202 to 0.260 depending on the model, which translates to 3 months of learning. Thus, the intervention effect on literacy scores equates to 3 months of literacy gains for children.

We examined the heterogeneity in the TbT effects on postintervention PALS-PreK score by calculating the estimated effect for subgroups.(Table 3) The intervention effect was evident for firstborn children (0.282, CI: 0.021 to 0.542, P = .034), 2-parent families (0.262, CI: 0.063 to 0.461, P = .010), English-speaking caregivers (0.238, CI: 0.003 to 0.473, P = .047), and children aged ≥48 months (0.436, CI: 0.116 to 0.755, P = .008). We found no difference by caregiver level of education.

The PCI scale was used to assess caregivers’ self-reported literacy interactions with their children (Table 4). The reading and total PCI scores did not reveal a significant change across unadjusted or adjusted analyses. The unadjusted intervention effect on postintervention PCI literacy activities was significant (1.96, CI: 0.167 to 3.76, P = .033); however, the intervention effect was not statistically significant when adjusted in Model 2 (1.25, CI: −0.329 to 2.83, P = .122) or Model 3 (1.22, CI: −0.328 to 2.77, P = .124).

Profound educational disparities are evident across socioeconomic groups at kindergarten entry. With this literacy-promoting texting program, we leveraged a pediatric care setting and increased literacy levels for a group of underresourced children by an average of 3 months. The effects varied across subgroups: firstborn and older children revealed more gains. These differential effects within our sample are not intended to guide clinical application. These findings are shared so that they may inform the design of future research into the effects of this program and parenting programs more generally. Overall, this is not one size fits all intervention.

Young children in the United States have greater access to health care than to early education because health care for young children is near universal. The Early and Periodic Screening, Diagnostic and Treatment benefit of Medicaid ensures all children (aged <18–21, state depending) are entitled to routine preventive services and medical treatment. Overall 49% of children aged 3–4 years are enrolled in preschool, with lower enrollment rates for those with parents who have lower educational levels,⁵¹  nearly 99% of children were hospital born,⁵²  and, at 5 years of age, 94% have received required immunizations for school entry.⁵³  As we have revealed, this cohort of low-income, publicly insured children represents a heterogeneous population. In our sample, those not in preschool had caregivers with significantly lower education and income and were more likely to identify as Hispanic and to speak Spanish at home. Educational interventions deployed from clinic can reach a distinct, more underresourced subset of families; this finding adds to our knowledge about the unique contribution the health sector can make toward addressing school readiness. More evidence-based, scalable programs are needed for younger and older children, as well as for those enrolled in preschool, to narrow ECE inequities.

TbT has the potential to move pediatrics to a higher level of impact. Past experience that scaled texting programs for parents through the education sector is informative. Within a few years after completing initial studies among preschool families in the San Francisco Unified School District, this program scaled quickly. The original research team has sent >40 million parenting texts to families in public preschools through the school districts in Florida, Georgia, New Jersey, Pennsylvania, Texas, and Wisconsin, among others. Large scale and rapid uptake were facilitated by the transformative ubiquity of personal cell phones, texting as a well-received modality of parent communication, and the intervention being inexpensive and scalable by design. However, although school districts have been instrumental in reaching families with children in preschool, they do not reach families with younger children or those not attending preschool. Health clinics can provide access to these families, but evidence of clear benefits for families and children are needed to justify moving to scale.

We used the PCI to assess program effects on parent self-reported behaviors and to better understand literacy improvement. The study was designed to measure the primary outcome of child literacy gains using PALS, rendering it underpowered to detect differences measured by PCI parent self-report. This choice was acceptable to investigators because early studies in preschools using teacher and parent reports detected positive impact on parenting behavior, as measured by PCI and teacher surveys, establishing the TbT causal mechanism.⁴⁵  Of note, at baseline, caregivers reported high reading levels, indicating that messaging to read to children may be reaching saturation, a fact to celebrate. However, caregivers who have heard they should be reading to their children may overreport reading, suggesting parent reports might not be accurate, given strong societal pressures. Tools that measure language directly by recording and counting number of words spoken⁵⁴  may better assess parental behavior.

TbT’s structure and scaffolding are based on extensive theoretical models that address caregiver challenges of low-resourced communities, including economic stress that requires adults to focus on short-term over long-term outcomes. For instance, a focus on having food for the week would displace a caregiver’s attention on longer term educational outcomes. TbT is designed to address these barriers, to break down educational processes into bite-size components that come via small doses every other day over a long period of time. As other nontexting clinic-based interventions emerge, such as individualized parent coaching,^55,56  library referrals,⁵⁷  electronic preschool enrollment,⁵⁸  or community referral coordinators,⁵⁹  sustained attention to caregiver needs and their day-to-day barriers will likely be important to overcoming the persistent differences in school readiness across groups.

The medical sector is increasingly motivated to address issues of poverty and its antecedent factors.^36,60  New advances in ECE are needed to address social-emotional learning, numeracy, literacy, and ways to support the development of newborn-to-3-year-olds. It requires humility to acknowledge that medical training does not prepare us to build effective ECE interventions. The role of convergence science,⁶¹  founded on trusting, transdisciplinary partnerships with schools of education, holds promise for transformative advances.

This study has a number of limitations. Northern California is racially diverse, with a significant Hispanic population, limiting generalizability. It remains to be seen whether rural families or urban, African American families would respond similarly, although early findings show promise.⁴⁵  Demographic data could not be gathered from those declining participation, so how they vary from participants is unknown. The PCI tool relied on parent self-report, which may have been impacted by social desirability bias and a ceiling effect, particularly for reading. Fidelity concerns are always warranted, but treatment fidelity is unusually strong in a text-message program. The caregivers we texted may have changed caregiving responsibilities or allowed others to use their phone. A final limitation is that the study is based on 1 measure of child development, and TbT can affect many aspects of child development beyond what PALS-PreK assesses.

This study revealed that a texting intervention sent from a clinical setting could reach a underresourced population of families, resulting in 3 months of literacy gains for young children. Further studies of clinic-based texting interventions exploring the causal factors linked to child outcomes are warranted. Leveraging the health sector’s unparalleled access to young children presents an opportunity to narrow educational disparities.

ECE

early childhood education

PALS-PreK

Phonological Awareness Literacy Screening Tool

PCI

Parent Child Interactivity Scale

RA

research assistant

TbT

TipsByText