[T]he data suggest that the major but by no means the sole factor determining the variability of attachment level or pocket depth changes is the nature of the local factors.

Haffajee AD, Socransky SS, Goodson JM, Lindhe J. Intraclass correlations of periodontal measurements. J Clin Periodontol 1985; 12: 216-24

The above quote is found in an old paper by leading periodontal scientists of the time who had been criticized by statisticians that their way to analyze hundreds of site-specific observations in a given patient as if they were independently sampled was fundamentally flawed. As the last author, Jan Lindhe, was the preeminent EiC of the then (and many years to come) most prestigious journal in dentistry, the paper could be read as a, well, defiant response which was supposed to solve the annoying issue once and for all.

The authors had re-analyzed their already published data of changes, after various forms of periodontal treatments, in key periodontal measurements, i.e. attachment levels (AL), pocket depth (PD) as well as gingival and plaque index scores. The data were subjected to comparing variance components and calculate intraclass correlation coefficents. Technically, in this scenario, the intraclass correlation coefficient (ICC) is the ratio of the variance attributable to subjects divided by the total of subject and error variances. By calculating it one gets an idea how strong the changes resembled each other at the patient level.

Ignoring the patient as random effect, statisticians had often demurred (seemingly without actually suggesting a suitable alternative) that, because observations were not independent as standard statistical analyses require, p-values would be inflated and hypothesis testing thus flawed.

As Haffajee et al. found out, ICCs were low or very low, in particular for probing measurements, AL (0.011 to 0.165) or PD (-0.009 to 0.178; read the later paper by Bruno Giraudeau on problems with negative ICCs).

The Constructive Advice of the Outstanding Editor-in-Chief of “The Journal“

I want to mention this early paper by Haffajee et al. (1985) as I had referred to it in 1985 when having submitted my series of some of my, as to my subjective judgment spectacular, early cases to “The Journal” (as young periodontists in Germany, and elsewhere, generally referred to the Journal of Clinical Periodontology) [1]. I wanted to make the point when writing to professor Jan Lindhe that my statistical analysis, ignoring patient effects, was justifiable in view of Haffajee’s results of low or very low ICCs.

Continue reading →