Reliability and validity of a photographic method for measuring facial hair density in men

Many studies have investigated hair removal or growth prevention treatments, but they often measure hair density using noninvasive methods that are subjective and qualitative.¹ Although photographic and digital hair-counting methods have been used, their reliability and validity remain unknown.² We describe a simple, noninvasive method of hair counting used in a hair growth prevention treatment trial and assess its reliability and validity.

Methods

The data are from the first 14 healthy men consecutively enrolled in a randomized, double-blinded, placebo-controlled trial of a topical agent for hair growth prevention. Eligible subjects were required to shave at least once daily to avoid a beard with hair length visible above the skin line and to have a baseline physician global assessment (PGA) score for hair density of 4 or 5 in the beard area. The PGA was developed by us for the larger clinical trial as a visual analog scale for rating hair density by overall impression (Figure).

Figure.

Protocol for hair density assessments

Subjects were randomized as to which side of their face would receive drug or placebo, which was then applied once daily after shaving to a treatment area within the beard region in a split-face design (Figure). The duration of active treatment was 6 or 8 weeks; subjects were assessed every 2 or 4 weeks for up to 8 to 16 weeks. Subjects did not shave for 48 hours prior to each visit so that they would have enough visible hair for assessment. At each visit, the PGA and digital photography of the treatment areas were performed (Figure). The study was approved by the University of Pennsylvania institutional review board.

Two of us (J.W. and J.M.S) independently counted hairs in all photographs to assess interrater reliability (Figure). Five months after the initial measurement, hairs were recounted in all photographs to assess test-retest reliability. We used the intraclass correlation coefficient (ICC) and Spearman ρ correlation to assess reliability. Construct validity was evaluated by comparing hair counts with respect to corresponding PGA ratings using the t-test. We conservatively estimated a sample size of 100 photographs with 85% power to detect an ICC of 0.6, assuming null ICC of 0.4 and α =0.05.

Results

The median age of the subjects was 28 years (interquartile range [IQR] 26–38 years). Eleven subjects were white (79%), and 3 were Asian (21%). All subjects had brown or black hair. A total of 130 photographs were obtained. Hair counts were approximately normally distributed, ranging from 2 to 391. The subject PGA scores were available for 114 photographs and ranged from 2 to 5 (median 4; IQR, 4-4). Test-retest reliability demonstrated an ICC of 0.90 (95% confidence interval [CI], 0.86–0.93) and a Spearman ρ of 0.88 (95% CI, 0.84–0.92). Interrater reliability demonstrated an ICC of 0.81 (95% CI, 0.74–0.86) and a Spearman ρ of 0.81 (95% CI, 0.75–0.87). In the validity analysis, we included only PGA scores for which there were at least 10 corresponding photographs. Photographs with a PGA score of 3 had a lower mean hair count (mean [SD] count, n = 195.0 [16.5]) than those with PGA score of 4 (mean [SD] count, n = 237.2 [5.8]) (p=0.003).

Comment

Our hair counting method demonstrates excellent interrater and intrarater reliability as well as construct validity based on its ability to discriminate categories of a PGA.³ In contrast to other methods, our approach does not require expensive or specialized equipment. It provides better quantification of hair changes than global assessment scales, which may be too qualitative for clinical trials.¹ Moreover, it is less tedious and labor intensive than the manual collection, counting, and weighing of hair.⁴ Although automated methods such as the Trichoscan have reported high reliability, fully automated approaches are hindered by imperfect algorithms, which can lead to inaccuracy.^{1, 5}

We recognize several limitations. First, hair diameter and length were not evaluated. Second, the camera was not mounted, and the skin in the treatment areas was not marked so as to guarantee the same exact evaluation distance and site every time. The generalizability of our results to areas with different hair density or to people with darker skin is unknown. Finally, additional studies are required to determine if this technique is responsive to true changes in hair density and to compare this method to other approaches such as digital photodermoscopy.

Nevertheless, our simple, noninvasive method of hair counting demonstrates excellent reliability and discrimination validity and deserves further evaluation as an assessment tool for hair removal or growth prevention studies.