Background There is evidence from existing literature of ethnic variations and clustering in adolescent health risk behaviours. However, it is not known whether there are ethnic variations in health risk behaviour clustering. We aimed to investigate ethnic variations in the clustering of health risk behaviours using the Determinants of young Adults Social wellbeing and Health study (DASH).

Methods Latent class analysis and multinomial logistic regression were used to investigate adolescent health risk behaviour clustering and ethnic variations in clustering in the MRC DASH study. Data were collected from 51 schools across 8 London boroughs. In 2005–06, 4785 pupils were followed up at 14–16 years old. Age, gender, ethnicity, substance use (SU) behaviours (current tobacco and alcohol, and lifetime illicit drug use), daily fruit and vegetables (FV), weekly physical activity (PA), and body mass index (BMI) were recorded.

Results A model with four latent classes was selected (entropy: 0.757). Classes could be characterised as high-SU/high-PA (n=232), high-SU/low-PA (n=811), low-SU/PA/high-FV (n=1471), and low-SU/low-PA/FV (n=2260).

Using the low-SU/PA/high-FV class as reference males were more likely to be in the high-SU/high-PA class (OR: 1.89; 95% CI: 1.39–2.57), and less likely to be in the high-SU/low-PA class (OR: 0.56; 95% CI: 0.47–0.67) than females.

Compared to White UK, Black Caribbean adolescents were less likely to be in the high-SU/high-PA (OR: 0.48; 95% CI: 0.30–0.76), or high-SU/low-PA (OR: 0.34; 95% CI: 0.26–0.45) classes. Black Africans were less likely to be in the high-SU/high-PA (OR: 0.18; 95% CI: 0.10–0.31), high-SU/low-PA (OR: 0.11; 95% CI: 0.08–0.15), or low-SU/low-PA/FV (OR: 0.55; 95% CI: 0.44–0.69) classes. Indians were less likely to be in the high-SU/high-PA (OR: 0.36; 95% CI: 0.19–0.71), or high-SU/low-PA (OR: 0.17; 95% CI: 0.11–0.26) classes. Pakistanis/Bangladeshis were less likely to be in the high-SU/high-PA (OR: 0.40; 95% CI: 0.23–0.68), high-SU/low-PA (OR: 0.08; 95% CI: 0.05–0.13), or low-SU/low-PA/FV (OR: 0.65; 95% CI: 0.50–0.85) classes.

Conclusion Latent class analysis is a valuable method for investigating ethnic variations in adolescent lifestyles. Ethnic minority adolescents tend to be in classes which are characterised by less unhealthy behaviour; patterns also vary between ethnic minority groups. Research needs to investigate risk and protective factors that may explain these ethnic variations to identify likely intervention targets and to inform public health policy.