The predictive validity of Bayley Scales of Infant and Toddler Development-III at 2 years for later general abilities: Findings from a rural, disadvantaged cohort in Pakistan

**Background:** Early childhood development (ECD) interventions are increasingly integrated into health programs in low- and middle-income countries (LMICs) to improve long-term cognitive outcomes. The Bayley Scales of Infant and Toddler Development (BSID) are widely used as a gold standard for assessing ECD in children under 3 years, but evidence on their predictive validity for later intellectual abilities in LMIC settings is scarce. This study aimed to evaluate the predictive validity of the BSID-III at 2 years for general intellectual abilities at ages 4 and 7–8 years in a rural, disadvantaged Pakistani cohort.

**Methods:** Data were drawn from a longitudinal follow-up of an intervention cohort (N = 1,298) in Naushahro Feroze, Sindh, Pakistan. At 2 years, children were assessed using the BSID-III (cognitive, language, and motor composite scores). At 4 years, general ability was measured using the Wechsler Preschool and Primary Scale of Intelligence (WPPSI-III), yielding Full-Scale IQ (FSIQ), Verbal IQ (VIQ), and Performance IQ (PIQ). At 7–8 years, the Fluid Reasoning Index (FRI) from the Wechsler Intelligence Scale for Children (WISC-V) was administered using two subtests (Matrix Reasoning and Figure Weights). Linear regression models were used to examine the proportion of variance (R²) in later IQ scores explained by BSID-III subscales. Stratified analyses were conducted by socioeconomic status, maternal literacy, child gender, and nutritional status.

**Key Results:** The cohort was highly disadvantaged: 60.6% of children were stunted, 43.2% were underweight, and only 31.6% of mothers were literate. Mean BSID-III composite scores ranged from 78.2 (SD 14.6) for cognitive to 88.7 (SD 17.0) for motor. Mean WPPSI-III FSIQ was 75.6 (SD 7.6), and mean WISC-V FRI was 60.9 (SD 10.4). Correlations between BSID-III subscales and WPPSI-III IQ scores were moderate and significant (r = 0.24 to 0.39, p < 0.001). The BSID-III language score showed the strongest correlation with WPPSI-III VIQ (r = 0.39, p < 0.001). In multiple regression models, the combined BSID-III subscales explained 15% of the variance in WPPSI-III FSIQ, 16% of VIQ, and 8% of PIQ at 4 years. For WISC-V FRI at 7–8 years, the BSID-III explained only 1% of the variance. When sociodemographic and anthropometric variables were added, the explained variance increased to 20% for WPPSI-III and 2% for WISC-V FRI. Stratified analyses showed higher predictive validity in low-risk groups (not stunted, higher SES, literate mothers) compared to high-risk groups. Mean differences in WPPSI-III FSIQ between the BSID-III reference group (>-1 SD) and the lowest category (<-2 SD) ranged from -4.87 to -5.28 across BSID-III subscales. For WISC-V FRI, mean differences ranged from -1.85 to -4.07.

**Clinical Implications:** The BSID-III at 24 months has only modest predictive validity for intellectual abilities at age 4 and very weak predictive validity for fluid reasoning at age 7–8 in this disadvantaged Pakistani cohort. Clinicians and researchers should use BSID-III scores with caution when attempting to forecast later cognitive outcomes, particularly in high-risk environments where multiple adversities may disrupt developmental trajectories. The language domain of the BSID-III showed the strongest associations with later IQ, suggesting early language skills may be particularly important for predicting later abilities. These findings underscore the need for culturally sensitive, longitudinally validated assessment tools in LMIC settings and highlight that early developmental assessments alone are insufficient to predict long-term cognitive outcomes in disadvantaged populations.