Validity of self-reported versus actual age in Nepali children and young people

Highlights • Self-reported age is a potential source of misclassification bias in International Surveys.• We compare objectively recorded age with self-reported age at mean age 11.5 years in 3943 children in rural Nepal.• There was high agreement between actual and self-reported age with an error rate of 7%.

Many international surveys of children and young people rely on age reported by parents or by young people. 1 Recording of births is not standardised, age may be reported approximately, and in cultures such as Nepal, South Asia, age may be reported in 'running' years, such that a child age 10 years will say they are 'running 11'. So self-reported age is potentially an important source of misclassification bias. There is a significant body of evidence examining the potential for misclassification bias in the report of health measures and health related behaviours, such as weight, 2 age of menopause 3,4 and smoking. 5 However, the accuracy of reported versus actual age has not been tested e there is no published study that compares actual with self-reported age in low income country settings.
The aim of this study was to assess the validity of selfreported age in a rural low income setting. Using data from a closed adolescent birth cohort study we compared objectively recorded age with self-reported age in full years in a group of 3943 children aged between recorded ages of 9.5 and 13.1 years (mean 11.5 years).
The mothers of the children and young people included in this cohort were part of a cluster randomized control trial of a community based intervention e women's participatory groups e in rural Makwanpur, Nepal. 6 All women who were pregnant during the trial period, Oct 2001 to Nov 2003, were invited to take part in the women's groups and around one third of this population chose to do so. The women's groups addressed issues around pregnancy, childbirth, and newborn and child health. The main outcomes, neonatal and maternal mortality, were assessed at four weeks postpartum by household interview for all women (both those who attended and those who did not attend women's groups). Maternal and neonatal mortality were significantly lower in intervention than control clusters. Date of birth was collected from the families by a field interviewer within four weeks of birth.
An average of 11.5 years later the first and only long term follow-up of this cohort was conducted (2014e2015). Surviving children who consented and were traceable at the time of follow-up underwent a face-to-face interview in their homes. During the interview children were asked to give their selfreported age. A family member, usually their mother, was present at all times during the interview. Self-reported age was recorded in full years.
The perinatal and long term follow-up data from this cohort were matched to create a closed adolescent birth cohort. Retention rate for surviving infants at four weeks of age to mean 11.5 years was 73%. An additional 3.2% of the original cohort died between trial completion and long term follow-up and 4.6% of the original cohort had missing age data (Fig. 1).
From the date of interview and the date of birth we calculated 'true' age in years e termed objectively recorded age. These were then categorized into complete years for example 10 years of age included all those with ages between 9.999 years and 10.999 years. Chi-squared and kappa tests were used to compare reported age groups with true age groups. Bland Altman plots were also used to assess measurement agreement.
The table shows around 90e95% agreement with significant correlation between actual and self-reported age (P < 0.001 for Chi 2 , Kappa value 0.89). Overall agreement between self-reported and recorded age was 93.1%. The overall error rate was therefore 6.9% (95% confidence interval, 6.1 to 7.7). The limits of agreement between each measure of age were À0.52e0.57 years. Self-reported age was greater than actual age by a mean difference of 0.02 years (95% CI 0.01 to 0.03). Pitman's Test of difference in variance was r ¼ 0.014 (P ¼ 0.394) Pitman's test examines for the equality of variance between two correlated samples, with the null hypothesis rejected here being that there is a significant difference in the variances of the two samples ( Table 1).
Strengths of our study are that we have reliable birth dates and interview dates in a resource poor setting. Limitations are that interviews were carried out with family members present and therefore we cannot be certain whether reported age was that from the child alone or from the family member present at interview. This would limit the generalizability of our p u b l i c h e a l t h 1 3 7 ( 2 0 1 6 ) 1 8 5 e1 8 7 findings to surveys that are carried out with interviews taking place solely with the child or young person. There was high agreement between reported and actual age. While it is reasonable to rely on reported age for national and international epidemiological surveys in low income settings like Nepal, our study did show an error rate of 6.9% (95% CI 6.1 to 7.7) which might influence overall estimates of nutritional status, immunization coverage, puberty scores and other child health variables.
Author statements