Pediatrics published an RCT of TIPP, a program to help pediatricians prevent injuries in children 0-2. Quick take: The main study finding - reduced child injuries - isn't reliable because this was a cluster RCT (4 sites were randomized) and the analysis didn't adjust for clustering.

​

• The study randomly assigned 4 academic medical centers, containing a total sample of 865 infant-parent dyads, to TIPP vs control. TIPP, designed by the American Academy of Pediatrics, helps pediatricians identify parental at-risk behavior and counsel/support parents in injury prevention.

• The study's randomization of just 4 medical centers is nowhere near large enough to produce highly-similar treatment and control groups - the essential condition for a valid RCT - because the 4 centers differ systematically in their patient populations, as shown here:  

• Given such cross-site differences, the study can't rule out the possibility that the difference in injury outcomes between treatment sites (Miami, UNC) and control sites (NYU, Vanderbilt) is due to underlying differences in their patients and/or physicians, rather than the TIPP program.

• The only reason the study was able to "find" statistically significant effects on injuries is that its analysis inappropriately ignored the clustering of patients within the 4 centers.