Correlation due to clustering in ratings of caries lesions uri icon
  • Objective: To estimate intraclass correlation (ICC) among ratings of depths of lesions restored by Dental Practice-Based Research Network (DPBRN; dentists, and its effect on study precision. This will inform both the design of future studies and appropriate analysis of current data. Methods: 229 DPBRN dentists recorded information on 9,890 restorations of previously-unrestored tooth surfaces in 5,810 patients. Correlation among clustered measurements reduces study precision. This design effect (DE) depends on the magnitude of the ICC and the number of observations per cluster. In this study, clustered observations occur due to inclusion of multiple patients per dentist and multiple restorations per patient. ICC and DE were estimated for ratings of pre- and post-operative lesion depths and for agreement between pre- and post-operative ratings. Estimates of correlation within dentists, with and without adjustment for regional means, and within patients were calculated. Results: The mean number of observations per dentist was 36. Unadjusted ICCs for clustering by dentist were 0.2151, 0.2050, and 0.1008 for pre-, post-, and agreement ratings, respectively. The corresponding DE values were 8.5, 8.2, and 4.5, respectively. Adjusting for differences in region means yielded ICC estimates of 0.1885 (DE = 7.6), 0.1454 (DE = 6.1) and 0.018 (DE = 1.6) for pre-, post-, and agreement, respectively. The mean number of ratings per patient was 1.4. Removing variability due to differences in dentist and region level means yielded ICCs of 0.3798 (DE = 1.2), 0.3219 (DE = 1.1), and 0.1595 (DE = 1.1) for pre-, post-, and agreement, respectively. Conclusion: Within-dentist correlation showed substantial DE which would reduce study precision relative to equal numbers of independent observations. Within-patient correlation, though of greater magnitude, showed smaller DE due to a small average number of ratings per patient. Support: U01-DE-16746, U01-DE-16747.

  • publication date
  • 2010
  • Research
  • Dental Caries
  • Research Design