Psoriasis Severity Assessment Combining Physician and Patient Reported Outcomes: The Optimal Psoriasis Assessment Tool

Patient Demographics

In this pooled analysis of 3866 patients, the average age was 45.5 years, 67.8% were male, and 92.6% were white. Mean baseline values for PASI and DLQI were 20.2 and 12.5, respectively. Baseline demographics and disease characteristics are shown in Table 1.

Table 1 Baseline demographics and disease characteristicsCorrelation Analyses

The patient-assessed measures of skin pain, itch, and PatGA were significantly correlated with both PASI and DLQI. Similarly, the correlation of BSA with PASI and DLQI was also established.

PASI and BSA were significantly (p < 0.001) correlated with each other at baseline (r = 0.759), week 4 (r = 0.745), week 8 (r = 0.804), and week 12 (r = 0.852) (Fig. 2a). There was a significant correlation between PASI and the product of BSA with worst scaling (r = 0.836–0.923), worst sum of severity (r = 0.857–0.939), worst erythema (r = 0.824–0.918), and worst thickness (r = 0.842–0.929) at baseline and weeks 4, 8, and 12 (Fig. 2a). Similarly, BSA alone was also significantly correlated with DLQI total score from baseline (r = 0.131; p < 0.001) to week 12 (r = 0.536; p < 0.001) (Fig. 2b).

Fig. 2figure2

a PASI correlations with simpler clinically assessed measures; b DLQI correlations with physician’s assessment measures

The correlations of DLQI with patient-assessed measures were higher than that with PASI for itch NRS and skin pain (itch NRS, 0.771 vs 0.695 at week 12; skin pain, 0.749 vs 0.635 at week 12), and similar for PatGA (PatGA, 0.748 vs 0.753) at week 12 (Fig. 3). The correlations at week 12 were higher than baseline for both PASI (skin pain, 0.154 vs 0.635; itch NRS, 0.128 vs 0.695; PatGA, 0.186 vs 0.753) and DLQI (skin pain, 0.507 vs 0.749; itch NRS, 0.490 vs 0.771; PatGA, 0.350 vs 0.748). Correlations between DLQI and the product of BSA with worst erythema, thickness, scaling or sum of severity at week 12 ranged from 0.584 to 0.596 (Fig. 2b).

Fig. 3figure3

PASI (and DLQI) correlations of patient-assessed measures with PASI as reference

Sensitivity and Specificity Analyses for PredictionModel with BSA and Itch

For the ordinary least-squares (OLS) regression model with BSA and itch, Table 2 shows sensitivity and specificity of the association between the percentage change in proxy PASI score with PASI 75 and PASI 90 at week 12. For PASI 75, a 73% change in proxy PASI score provides the maximum association (based on Youden index), whereas an 83% change in proxy PASI score had the maximum association for PASI 90.

Table 2 Percentage improvement in proxy PASI score

Sensitivity, specificity, positive predictive value, and negative predictive value were very high (greater than or equal to 80%) for both PASI 75 and PASI 90. PASI 75 had a sensitivity of 87.5% whereas PASI 90 had a sensitivity of 88.0%.

Model with BSA and PatGA

For the OLS model with BSA and PatGA, Table 2 shows sensitivity and specificity of the association between the percentage change in proxy PASI score with PASI 75 and PASI 90 at week 12. For PASI 75, a 71% change in proxy PASI score provides the maximum association (based on Youden index), whereas an 89% change in proxy PASI score had the maximum association for PASI 90.

Sensitivity, specificity, positive predictive value, and negative predictive value were very high (greater than or equal to 81%) for both PASI 75 and PASI 90. PASI 75 had a sensitivity of 88.6% whereas PASI 90 had a sensitivity of 81.4%.

Concordance/Discordance

The concordance summaries showed high concordance rates for PASI 75 and PASI 90, as shown for the models with PatGA (77.8% and 60.2%) and itch (74.4% and 55.1%) for PASI 75 and PASI 90, respectively, as shown in Fig. 4.

Fig. 4figure4

Concordance rates by residual BSA severity at week 12 in proxy BSA (all treatment arms)

Models with BSA and itch, and BSA and PatGA were similar in terms of percentage improvement in proxy PASI score and sensitivity (Table 2). Furthermore, the most severe patients at baseline who had good responses (PASI 75 and PASI 90) after 12 weeks often continued to have BSA involvement greater than 10%, suggesting that relying on BSA changes alone does not fully capture the benefit or amount of clinical improvement.

In assessing these models, the scatterplots (Fig. 5) visually displayed the alignment of the proxy PASI versus PASI. The concordance summaries showed very high concordance rates for PASI 75 and PASI 90, especially for the models with PatGA and itch.

Fig. 5figure5

Predicted vs observed PASI scores

Model Comparison Based on Regression Analysis

At week 12, the correlations between PASI and BSA combined with patient assessments [PatGA (\(\mathrm(\sqrt^})\) = 0.904), itch NRS (\(\mathrm\) = 0.898), and skin pain (\(\mathrm\) = 0.890)] were higher than between PASI and BSA alone [\(\mathrm\) = 0.852 (Fig. 6)].

Fig. 6figure6

PASI (and DLQI) correlations with combined clinical and patient assessments at week 12

In case of DLQI total score, the correlations with BSA combined with the same measures [PatGA (\(\mathrm\) = 0.756), itch NRS (\(\mathrm\) = 0.785), and skin pain (\(\mathrm\) = 0.775)] were stronger than with BSA alone (\(\mathrm\) = 0.536) or with PASI (Corr = 0.662) at week 12. The proxy PASI versus actual PASI results were plotted for each of the two-parameter models. In addition, for each model, the concordance rates were summarized for the proxy measure versus PASI.

For the overall population at week 12, of the 18 models tested for PASI (Supplementary Table 1), the results for the MAE, PRESS, and RMSE from overall datasets ranged from 2.12 to 4.02, 12.61 to 34.62, and 3.52 to 5.87, respectively. MAE, sqrt(PRESS), and RMSE represented percentage errors of up to 5.6%, 8.2%, and 8.2%, respectively, for the overall PASI range. In the PASI models, out of the three patient assessment variables, those models with PatGA had the best performance with respect to common model selection criteria (AIC, BIC, or RMSE in validation dataset), followed by itch and skin pain. In the DLQI models (Supplementary Table 2), MAE, PRESS, and RMSE from overall datasets ranged from 2.21 to 3.62, 11.98 to 26.42, and 3.42 to 5.14, respectively. Percentage errors for MAE, sqrt(PRESS), and RMSE were up to 12.1%, 17.1%, and 17.1%, respectively, for the overall DLQI range. The DLQI model with itch had the best performance, followed by skin pain and PatGA.

As shown in Supplementary Tables 1 and 2, the most complicated models with BSA, itch, PatGA, skin pain, and their interaction terms have the smallest AIC, BIC, or RMSE for the validation dataset. A summary of estimated coefficients of selected models is presented in Tables 3 and 4. However, the best two-term models [i.e., BSA + PatGA (R2 = 0.82), and BSA + itch (R2 = 0.81) for PASI and BSA + itch (R2 = 0.62) for DLQI] or three-term models [i.e., BSA + PatGA + BSA × PatGA (R2 = 0.84) for PASI and BSA + itch + skin pain (R2 = 0.65) for DLQI] are still preferred, as they showed minor differences in R2 and RMSE in comparison to the most complicated models.

Table 3 Summary of coefficients (estimates) of selected models (without interaction and without higher order terms) for PASI at week 12Table 4 Summary of coefficients (estimates) of selected models (without interaction and without higher-order terms) for DLQI at week 12

A sample digital prototype of OPAT is presented in Fig. 7.

Fig. 7figure7

Digital prototype of OPAT

留言 (0)

沒有登入
gif