Dermatol Ther (Heidelb) (2021) 11:1291–1304 https://doi.org/10.1007/s13555-021-00550-8 ORIGINAL RESEARCH Deterioration of Health-Related Quality of Life After Withdrawal of Risankizumab Treatment in Patients with Moderate-to-Severe Plaque Psoriasis: A Machine Learning Predictive Model Kim A. Papp . Ahmed M. Soliman . Nicolae Done . Christopher Carley . Esteban Lemus Wirtz . Luis Puig Received: March 26, 2021 / Accepted: May 6, 2021 / Published online: May 21, 2021 Ó The Author(s) 2021 ABSTRACT 16 in IMMhance, a machine learning predictive model for DLQI was developed. Introduction: Risankizumab has demonstrated Methods: A machine learning model for DLQI efficacy in treating moderate-to-severe psoriasis. was fitted using repeated measures data from The phase-3 IMMhance trial (NCT02672852) three phase-3 trials (NCT02684370, examined the effect of continuing versus with- NCT02684357, NCT02694523) (pooled drawing from risankizumab treatment on pso- N = 1602). An elastic-net algorithm performed riasis severity, including the Psoriasis Area and automated variable selection among candidate Severity Index (PASI) and static Physician Glo- predictors including concurrent PASI and sPGA, bal Assessment (sPGA). However, the effect of demographics, and interaction terms. The withdrawal on health-related quality of life machine learning model was used to predict (HRQL) was not assessed. Therefore, this study DLQI at weeks 28–104 of IMMhance among was conducted to evaluate the impact of risan- patients re-randomized to continue (N = 111) or kizumab withdrawal on HRQL measured by the withdraw from (N = 225) risankizumab after Dermatology Life Quality Index (DLQI). achieving response (sPGA = 0/1) at week 28. Because DLQI was not measured beyond week Results: The machine learning predictive model demonstrated good statistical fit during tenfold cross-validation and external validation Supplementary Information The online version against observed DLQI at weeks 0–16 of contains supplementary material available at https:// doi.org/10.1007/s13555-021-00550-8. IMMhance (N = 507). Predicted improvements in DLQI from baseline were lower in the with- K. A. Papp (&) drawal versus the continuation cohort (mean K Papp Clinical Research and Probity Medical DLQI change at week 104, -5.9 versus -11.5, Research, Waterloo, ON, Canada difference [95% CI] = 5.6 [4.1, 7.3]). Predicted e-mail: kapapp@probitymedical.com DLQI deteriorated more extensively than PASI A. M. Soliman (49.7% versus 36.4%) after treatment AbbVie, Inc., North Chicago, IL, USA withdrawal. Conclusions: The predicted DLQI score deteri- N. Done  C. Carley  E. Lemus Wirtz Analysis Group, Inc., Boston, MA, USA orated more rapidly after risankizumab with- drawal than the PASI score, an objective L. Puig measure of disease. These findings suggest that Department of Dermatology, Hospital de la Santa the deterioration in HRQL reflects more sub- Creu i Sant Pau, Universitat Autònoma de Barcelona, Barcelona, Spain stantial impacts after risankizumab 1292 Dermatol Ther (Heidelb) (2021) 11:1291–1304 discontinuation than those measured by PASI DIGITAL FEATURES only. This article is published with digital features, Keywords: DLQI; Machine learning; PASI; including a summary slide, to facilitate under- Psoriasis; Quality of life; Risankizumab standing of the article. To view digital features for this article, go to https://doi.org/10.6084/ m9.figshare.14541864. Key Summary Points Why carry out this study? INTRODUCTION Although the efficacy of risankizumab in the treatment of moderate-to-severe Plaque psoriasis is a chronic inflammatory skin psoriasis has been demonstrated in a condition, characterized by pruritic and scaly recent phase 3 clinical trial, the impact of lesions (plaques) throughout the patient’s body its withdrawal on patients’ health-related [1]. Patients with plaque psoriasis experience quality of life (HRQL) has not yet been debilitating pain, stigmatization [2], and psy- assessed. chosocial impairments [3, 4] due to the appearance of lesions, and are susceptible to This study was conducted to gain insight multiple comorbidities (e.g., metabolic syn- regarding the impact of risankizumab drome, cardiovascular disease) [5]. This has withdrawal on HRQL among patients with resulted in a substantial decrease in patients’ moderate-to-severe psoriasis from health-related quality of life (HRQL) [6–9], IMMhance using the Dermatology Life which typically correlates with disease severity Quality Index (DLQI); however, due to the (e.g., mild, moderate, severe) [5]. absence of DLQI collected beyond week Although there is no cure for psoriasis, 16 in clinical trials, a machine learning treatments are available to assist in providing predictive model was developed to assess sustained clearance of lesions, particularly the DLQI among patients with moderate- among patients with moderate-to-severe plaque to-severe psoriasis. psoriasis [10]. Several clinician-reported out- What was learned from the study? come measures are used to evaluate the mag- nitude of the disease burden and the The DLQI scores calculated using the effectiveness of therapies in clinical studies and machine learning predictive model real-world practice [11]. The Psoriasis Area and showed that patients who withdrew from Severity Index (PASI) is a commonly used risankizumab experienced more extensive instrument for measuring psoriasis severity and deterioration in their HRQL compared evaluating treatment response. PASI scores with patients who continuously remained measure psoriasis symptom involvement of four on treatment. body sections: head, upper extremities, trunk, and lower extremities [12]. The static Physician When compared with the Psoriasis Area Global Assessment (sPGA) is a widely used, and Severity Index (PASI), a commonly clinician-reported instrument for measuring used objective measure of clinical severity, psoriasis severity based on induration, scaling, the DLQI score showed a larger relative and erythema, with scores from 0 (clear) to 5 decline, suggesting that this measure of (severe) [13]. Although clinician-reported HRQL is affected by risankizumab instruments can provide objective assessments withdrawal more extensively than clinical of symptom control, assessments of treatment measures would suggest. effectiveness must also include the patient per- spective, particularly during clinical trials used to inform treatment selection and optimization in real-world practice. This requires the use of Dermatol Ther (Heidelb) (2021) 11:1291–1304 1293 instruments that incorporate patient-reported UltIMMa-1 and UltIMMa-2 Trials outcomes (PROs) [14, 15]. The Dermatology Patients in the UltIMMa-1 and UltIMMa-2 trials Quality Index (DLQI) is a PRO tool used to were randomized 3:1:1 at baseline to risankizu- assess HRQL across six domains (patient’s per- mab (150 mg at weeks 0 and 4 followed by every ception of symptoms and feelings, daily activi- 12 weeks), ustekinumab (45 or 90 mg at weeks 0 ties, leisure, work and school, interpersonal and 4 followed by every 12 weeks), or placebo relationships, and treatment) [16]. (weeks 0 and 4 followed by cross-over to risan- Risankizumab is a fully humanized mono- kizumab 150 mg every 12 weeks after an initial clonal antibody with a high affinity for the p19 16-week double-blind period)[19]. DLQI was component of interleukin (IL)-23, a cytokine collected at weeks 0, 12, 16, and 52. Candidate that plays a key role in the pathogenesis of predictor variables were collected at all visits psoriasis [17]. Recently, data from four large, through week 52. multicenter, phase 3 randomized controlled trials, including UltIMMa-1 (NCT02684370), IMMvent Trial UltIMMa-2 (NCT02684357), IMMvent Patients in the IMMvent trial were randomized (NCT02694523), and IMMhance 1:1 at baseline to risankizumab (150 mg at (NCT02672852), showed significantly greater weeks 0 and 4 followed by every 12 weeks) or efficacy with risankizumab versus placebo adalimumab (90 mg at week 0, followed by [18–20], ustekinumab [19], and adalimumab 40 mg at week 1 and every 2 weeks there- [20] as measured by relative improvements from after)[20]. At week 16, patients initially ran- baseline in PASI and DLQI. The IMMhance trial domized to adalimumab were switched to also investigated the impact of randomized risankizumab if \ PASI50, remained on adali- withdrawal from risankizumab among patients mumab if C PASI90, or were re-randomized 1:1 who achieved an initial response at week 28 to either risankizumab or adalimumab if C [18]. Data from IMMhance showed that con- PASI50 and \ PASI90. DLQI was collected at tinuous treatment provided significantly better weeks 0, 16, and 44. Candidate predictor vari- long-term disease control than withdrawal (at ables were collected at all visits through week week 52: 87% versus 61%, p \ 0.001) [21]. 44. However, in IMMhance, DLQI was only col- lected through week 16, and thus the corre- IMMhance Trial sponding impact of treatment withdrawal on Patients in the IMMhance trial were random- HRQL could not be directly assessed. To cir- ized 4:1 at baseline to risankizumab 150 mg or cumvent the limitation of outcome availability placebo [18]. Patients received a dose of study only for a limited number of patient visits, the medication at week 0, week 4, and every current study assessed the deterioration of 12 weeks thereafter. At week 16, patients ran- HRQL following risankizumab treatment with- domized to placebo started receiving active drawal using a machine learning predictive treatment with risankizumab 150 mg every model to assess DLQI among patients in the 12 weeks and continued until the end of the IMMhance trial. treatment period. At week 28, patients ran- domized to risankizumab who achieved an METHODS sPGA score of 0 or 1 were re-randomized to either risankizumab or placebo. Starting from Data Sources week 32, week 28 responders who experienced relapse (sPGA C 3) were switched to open-label risankizumab 150 mg every 12 weeks. DLQI was The data used in this study were from four collected at weeks 0, 12, and 16. Candidate phase-3 randomized, double-blind controlled, predictor variables were collected at all visits trials (UltIMMa-1, UltIMMa-2, IMMvent, and through week 104. IMMhance) of risankizumab for the treatment of moderate-to-severe plaque psoriasis (Fig. 1). 1294 Dermatol Ther (Heidelb) (2021) 11:1291–1304 Fig. 1 Data sources and patient flowchart in the clinical were switched to open-label risankizumab 150 mg. ADA trials included in the analysis. At week 28 of IMMhance, adalimumab, Q2W every 2 weeks, Q12W every 12 weeks, patients originally randomized to risankizumab who RZB risankizumab, UST ustekinumab, Wk week achieved a sPGA score of 0 or 1 were re-randomized to either risankizumab or placebo. Starting from week 32, week 28 responders who experienced relapse (sPGA C 3) Inclusion Criteria with human participants performed by any of the authors. The trials included in this study Model development included all patients in the were approved by each site’s institutional intention-to-treat populations of UltIMMa-1, review board and were performed in accordance UltIMMa-2, and IMMvent who had visits with with the Helsinki Declaration of 1964, and its concurrently available DLQI and predictor later amendments. variable measurements. External model valida- tion included all patients in the intention-to- Outcome Measure and Candidate treat population of the IMMhance trial who had Predictors visits with concurrently available DLQI and predictor variable measurements at weeks 0, 12, The outcome of interest was the total DLQI or 16. In the score prediction phase, DLQI score, which served as the dependent variable in scores were predicted among patients originally the predictive regression models. randomized to risankizumab in IMMhance who Candidate predictors including the PASI achieved sPGA of 0 or 1 at week 28 and were total score, PASI body section subscores, and the consequently re-randomized to continue risan- sPGA score were considered for inclusion in the kizumab or switch to placebo. The treatment prediction model of DLQI as concurrent mea- cohort was defined according to the intention- sures of psoriasis severity. Baseline candidate to-treat re-randomization group. DLQI scores predictors included demographics (age, sex, and were predicted for patient visits with concur- race), weight, duration of psoriasis, and history rently available predictor variable of use of systemic biologic therapy. Addition- measurements. ally, several squared terms (age2, weight2, dis- ease duration2, sPGA2, total PASI2) and Compliance with Ethics Guidelines interaction terms (PASI 9 age, PASI 9 weight, PASI 9 sex, sPGA 9 age, sPGA 9 weight, This article is based on previously conducted sPGA 9 sex) were considered as candidate studies and does not contain any new studies predictors. Dermatol Ther (Heidelb) (2021) 11:1291–1304 1295 Baseline Characteristics B (weeks 28 ?) of the IMMhance trial, the mean change in DLQI score from baseline was mod- Baseline characteristics were summarized for eled in each re-randomization group based on the intention-to-treat populations in UltIMMa- the difference between DLQI scores predicted at 1, UltIMMa-2, IMMvent, and IMMhance, and the visit versus the score observed at baseline. A for each re-randomized cohort in IMMhance. bootstrapping approach was used to obtain 95% Patient characteristics assessed included DLQI confidence intervals (CIs) around the mean and all candidate predictors. Means and stan- change from baseline in DLQI in each re-ran- dard deviations were summarized for continu- domization group and around the between- ous variables; counts and percentages were group difference at each visit. Percent mainte- summarized for categorical variables. Statistical nance of score improvement from baseline to comparisons between re-randomized cohorts in week 28 was assessed for predicted DLQI scores the IMMhance trial were conducted using one- and observed PASI in each re-randomization way analysis of variance for continuous vari- group. For patients who switched to open-label ables and v2 tests for dichotomous variables. risankizumab because of relapse, or who were lost to follow-up, last observation carried for- Model Development and Validation ward (LOCF) was used to impute PASI and sPGA scores at subsequent visits. Additional method- ological details are provided in Appendix 1 of Repeated-measures linear regression models for the electronic supplementary material. DLQI score prediction were developed with pooled data from UltIMMa-1, UltIMMa-2, and IMMvent, using cross-validation to prevent RESULTS model overfitting. In addition to several model specifications with progressively expanded pre- Baseline Characteristics dictor subsets, machine learning algorithms including elastic net and least absolute shrink- A total of 2109 subjects were included in the age and selection operator (LASSO) regression samples used for model training and validation were used to automate variable selection from (UltIMMa-1 and UltIMMa-2: 997; IMMvent: among the full set of candidate predictors. The 605; IMMhance: 507), and 336 subjects were best-fitting models were further validated in the included in the sample used for DLQI predic- IMMhance population using all patient visits tion (IMMhance re-randomized placebo: 225; with concurrently available DLQI scores (at risankizumab: 111). Baseline characteristics weeks 0, 12, and 16) and predictor variables. were well balanced across the intention-to-treat The predictive performance of all models was populations of the clinical trials used in the assessed using the cross-validated mean abso- model development and validation phase lute error (MAE) and the coefficient of deter- (Table 1). The mean age across cohorts ranged mination (R2). between 46.2 and 49.3 years, and between 25.2% and 30.7% of patients were female. The DLQI Prediction and Calculation of DLQI mean DLQI score ranged between 13.0 and and PASI Deterioration 13.7, and the total PASI score ranged between 19.7 and 20.2. Similarly, baseline characteristics Longitudinal analyses of predicted DLQI scores were balanced across the re-randomized cohorts were conducted to examine the deterioration of used for longitudinal DLQI score predictions. HRQL associated with risankizumab treatment withdrawal after achievement of response in the Model Development and Validation IMMhance trial. The treatment cohort [risanki- zumab versus placebo, (i.e., withdrawal)] was The prediction models for DLQI demonstrated defined according to intention-to-treat re-ran- significant, positive associations between DLQI domization groups. For each study visit in part 1296 Dermatol Ther (Heidelb) (2021) 11:1291–1304 Table 1 Baseline characteristics for patients whose data were used in the study Sample used for model training and validation Sample used for DLQI prediction (IMMhance re- randomized) UltIMMa-1 & IMMvent IMMhance PBO RZB p- UltIMMa-2 patients patients value N = 997 N = 605 N = 507 N = 225 N = 111 Demographics and clinical characteristics Age (years), mean (SD) 47.4 ± 13.7 46.2 ± 13.5 49.2 ± 13.3 49.3 ± 13.1 48.2 ± 13.4 0.457 Female, n (%) 300 (30.1%) 183 (30.2%) 151 (29.8%) 69 (30.7%) 28 (25.2%) 0.301 White race, n (%) 778 (78.0%) 508 (84.0%) 402 (79.3%) 177 (78.7%) 82 (73.9%) 0.326 Weight (kg), mean (SD) 90.1 ± 21.9 90.1 ± 23.9 92.0 ± 22.9 91.0 ± 22.2 91.3 ± 23.7 0.897 Duration of psoriasis (years), 18.1 ± 12.5 18.2 ± 12.3 19.9 ± 13.0 20.4 ± 13.4 19.4 ± 13.1 0.505 mean (SD) Prior biologic systemic therapy, 377 (37.8%) 229 (37.9%) 281 (55.4%) 125 (55.6%) 57 (51.4%) 0.467 n (%) Measure of disease severity, mean (SD) DLQI 13.0 ± 7.0 13.7 ± 7.2 13.4 ± 7.1 13.0 ± 7.2 13.0 ± 6.9 0.992 sPGA 3.2 ± 0.4 3.2 ± 0.4 3.2 ± 0.4 3.2 ± 0.4 3.2 ± 0.4 0.302 PASI score Total PASI score 20.1 ± 7.4 19.8 ± 7.5 20.2 ± 8.1 19.7 ± 7.3 20.2 ± 8.9 0.526 Subtotal (area score 9 lesion score) Head 16.3 ± 11.9 16.9 ± 11.8 17.0 ± 11.2 16.7 ± 10.3 18.8 ± 13.1 0.113 Trunk 18.1 ± 10.3 17.9 ± 9.8 18.5 ± 10.7 18.2 ± 9.5 18.5 ± 11.8 0.759 Upper extremities 19.5 ± 9.1 18.5 ± 9.1 19.4 ± 9.1 19.3 ± 9.0 18.9 ± 9.0 0.724 Lower extremities 22.9 ± 9.7 22.7 ± 10.1 22.6 ± 10.5 21.7 ± 9.7 22.6 ± 11.4 0.485 DLQI Dermatology Life Quality Index, PASI Psoriasis Area and Severity Index, PBO placebo, RZB risankizumab, SD standard deviation, sPGA static Physician’s Global Assessment and PASI across all specifications tested (Sup- IMMhance trial mirrored the observed DLQI plemental Table S1). The saturated model and scores closely at weeks 0, 12, and 16, further elastic-net selected model demonstrated the demonstrating the high out-of-sample predic- best predictive performance out of the models tive performance of the elastic net model tested, as measured by R2 and MAE. Due to the (Fig. 2a). Moreover, the predicted DLQI scores high predictive performance and relative parsi- closely mirrored the observed scores for patients mony, the elastic net model was chosen for the in the UltIMMa-1/2 and IMMvent trials who prediction of longitudinal DLQI scores among achieved PASI100 response at week 16 but then re-randomized patients in part B of the lost this response at week 52 or 44 (Fig. 2b), IMMhance trial. confirming the high predictive performance of The predicted DLQI scores among patients the elastic net model among patients who who were re-randomized at week 28 in the Dermatol Ther (Heidelb) (2021) 11:1291–1304 1297 Fig. 2 a Mean predicted versus observed DLQI scores for response at week 16 and then lost PASI 100 response at weeks 0–16 among patients re-randomized at week 28 in week 44 or 52. DLQI Dermatology Life Quality Index, the IMMhance trial; b mean predicted versus observed PASI Psoriasis Area and Severity Index DLQI scores in the UltIMMa-1, UltIMMa-2, and IMMvent trials among patients who achieved PASI 100 respond to therapy and then experience a the change in DLQI from baseline between deterioration of response. cohorts at week 52 [difference-in-differ- ence = 1.7; 95% CI = (0.2, 3.4)]. Beyond week Predicted Change in DLQI 52, increasingly larger mean changes in DLQI scores were observed in the withdrawal cohort Analysis of predicted DLQI scores using the relative to the risankizumab cohort. These dif- elastic net model revealed significantly greater ferences were significant at each visit beyond deterioration in the cohort re-randomized to week 52. Raw predicted DLQI scores among re- placebo relative to the cohort re-randomized to randomized patients are shown in Supplemen- continue risankizumab treatment (Fig. 3a). tal Fig. S1. There was a statistically significant difference in 1298 Dermatol Ther (Heidelb) (2021) 11:1291–1304 Fig. 3 a Change in predicted DLQI from baseline among PASI scores were assessed among patients originally re-randomized patients in the IMMhance trial. DLQI randomized to risankizumab who were consequently re- scores were predicted among patients originally random- randomized to continue risankizumab or switch to placebo ized to risankizumab who were consequently re-random- in part B (weeks 28 ?) of the IMMhance trial. An LOCF ized to continue risankizumab or switch to placebo in part method was used to carry forward measurements for visits B (weeks 28 ?) of the IMMhance trial. An LOCF with missing measures. CI confidence interval, DLQI method was used to carry forward measurements for visits Dermatology Life Quality Index, LOCF last observation with missing measures; b change in PASI from baseline carried forward, PBO placebo, RZB risankizumab among re-randomized patients in the IMMhance trial. Maintenance of DLQI and PASI Score change at week 104, -5.9 versus -11.5, differ- Improvement ence, 95% CI = 5.6 (4.1, 7.3)] (Fig. 3a). For patients in the withdrawal cohort, the DLQI The cohort re-randomized to placebo showed score improvement achieved from baseline to deterioration in observed PASI scores and pre- week 28 deteriorated more rapidly after with- dicted DLQI scores over time, while the cohort drawal compared with the deterioration in PASI re-randomized to continue risankizumab treat- score improvement (Fig. 4). At the end of fol- ment had near-complete maintenance of both low-up (week 104), DLQI score improvement scores (Fig. 3). Predicted improvements in DLQI from baseline to week 28 had deteriorated by from baseline were lower in the withdrawal 49.7%; PASI score improvement at week 28 versus the continuation cohort [mean DLQI deteriorated by 36.4% by week 104. The Dermatol Ther (Heidelb) (2021) 11:1291–1304 1299 Fig. 4 Maintenance of DLQI and PASI score improve- indicates a return to baseline score. The vertical line at ment following treatment re-randomization in the week 40 indicates the first visit when the percent IMMhance trial. Maintenance of score improvement from deterioration difference in DLQI versus PASI becomes week 28 was calculated as 100%—[100% 9 (change in statistically significant. DLQI Dermatology Life Quality score from baseline to week 28 - change in score from Index, LOCF last observation carried forward, PASI baseline to week x)/(change in score from baseline to week Psoriasis Area and Severity Score, PBO placebo, RZB 28)]. Maintenance of 100% indicates complete mainte- risankizumab nance of week 28 improvement; maintenance of 0% difference in deterioration of DLQI versus PASI higher rate compared with the PASI score fol- among patients who withdrew from risankizu- lowing treatment discontinuation. Despite the mab was statistically significantly different by significant durability of patients’ response to week 40 [difference (95% CI) = -1.6% (-2.9%, risankizumab following withdrawal, the deteri- -0.2%)], and in all weeks thereafter (Fig. 4). oration of HRQL is more rapid. These findings Changes in sPGA from baseline are shown in align with previous studies, which show that, Supplemental Figure S2. regardless of the reason for treatment gaps, treatment discontinuation negatively affects patients’ HRQL, particularly those responsive to DISCUSSION therapy. Results from a post hoc subanalysis of the double-blind, phase 3 REVEAL trial reported Achieving skin clearance is a meaningful a twofold greater deterioration of HRQL mea- improvement to patients with moderate-to- sured by DLQI compared with PASI following severe psoriasis [22]. Due to the chronic nature withdrawal of adalimumab [30]. Although the of the disease, patients with psoriasis require results do not allow the identification of the ongoing treatment to adequately manage cause for the disproportionate worsening of symptoms [23]. Although guidelines recom- DLQI, previous research suggests that psycho- mend continuous therapy for patients with logical impacts of disease worsening may be psoriasis who are treated with biologic agents responsible for this pattern [30, 31]. Given the [24, 25], gaps in treatment are common in importance in identifying the impact of chan- clinical practice [26, 27]. Reasons for treatment ges in HRQL to guide treatment optimization, interruption vary but can include lack of additional studies are warranted to further meaningful improvement, adverse events, and/ assess the impact of treatment withdrawal or patient preference to avoid continuous among patients with psoriasis. treatment [28, 29]. The present study demon- Methodologically, this study used several strated that the predicted HRQL worsened at a regression and machine learning model 1300 Dermatol Ther (Heidelb) (2021) 11:1291–1304 specifications based on patient characteristics and HRQL as measured by EQ-5D-3L [40], (e.g., age, disease duration, treatment history) which also found a nonlinear association and well-established measures of disease sever- between the two measures. In that analysis, ity to achieve robust performance in predicting increases in PASI close to zero were shown to DLQI scores and thus overcome the challenge of have a larger impact on EQ-5D-3L compared unmeasured scores. The cross-validation proce- with an equally sized increase in PASI at the dure helped protect against overfitting, while higher end of the scale, as determined by the the additional model testing with IMMhance inclusion of a PASI squared term as a significant data at weeks 0, 12, and 16 indicated high out- predictor of EQ-5D-3L. of-sample predictive ability. Results of the pre- sent study showed that the elastic net model Limitations achieved the best trade-off between predictive accuracy and model parsimony, highlighting This study has several limitations. First, as with the value of machine learning approaches in all trial-based analyses, the results may not be selecting relevant predictors among multiple generalizable to a real-world population, as the highly correlated measures. study population comprised a group of patients Overall, the findings from the machine selected on the basis of specific enrollment cri- learning procedure align with results from prior teria and followed under well-controlled con- studies that analyzed determinants of HRQL as ditions. Second, despite the high predictive assessed using DLQI [32–38]. Several studies to performance of the selected model, the exclu- date have reported a strong relationship sion of other variables not measured in the trial between PASI and DLQI, in diverse patient or unavailable at all time points may contribute populations and various settings worldwide to prediction error at the individual patient [32–38]. Moreover, this is the first study to level. The model closely predicted aggregate- develop a model for out-of-sample prediction of level average DLQI scores during external vali- DLQI based on a diverse set of patient and dis- dation, but does not fully explain heterogeneity ease characteristics in psoriasis, and explains a in DLQI scores across different patients. Third, higher proportion of the variation than a pre- this study assumed that achieving a given level vious multivariate model [33]. Based on the of clinical symptom control (as measured by models presented, sPGA was predictive of DLQI PASI and sPGA scores) has the same HRQL independently of PASI. This is consistent with benefits regardless of the specific treatment previous research showing that, although sPGA strategy that resulted in that symptom control. and PASI correlate well, the correlation is However, patients may experience different weaker at the ends of the distribution, which types and rates of adverse events depending on indicates that the two measures may provide treatment, which could also contribute to independent information about disease severity heterogeneity in the HRQL benefits of different [39]. The residual predictive ability of the head treatment strategies. Lastly, inherent limita- subscore of PASI was also consistent with a prior tions exist for the assessments of interest. For study that demonstrated that patients with example, limitations with the DLQI include psoriasis with visible psoriatic lesions have sig- inadequate measurement of mild illness and nificantly worse HRQL compared with those differential item functioning by disease, age, with nonvisible lesions [33]. and sex [41, 42], whereas limitations of PASI Finally, the preferred model specification include poor sensitivity, nonlinear scale, lack of suggests that the relationship between PASI and correspondence to clinical severity categories DLQI is nonlinear, as indicated by the selection [43], and dependence on the experience level of of the PASI squared terms as statistically signif- the clinician performing the measurement [44]. icant predictors of DLQI scores in addition to Moreover, lower PASI values are less clinically the linear PASI total score term. This is consis- meaningful, thus precluding valid comparisons tent with a previous analysis of the relationship across the entire spectrum of PASI scores [30]. between disease severity as measured by PASI Dermatol Ther (Heidelb) (2021) 11:1291–1304 1301 CONCLUSIONS Myers Squibb, Celgene, Coherus, Dermira, Eli Lilly, Forward Pharma, Galderma, Genentech, Predicted DLQI scores revealed an association GlaxoSmithKline, Janssen, Kyowa-Hakko Kirin, between risankizumab treatment withdrawal LEO Pharma, MedImmune, Merck-Serono, and HRQL deterioration, as evidenced by a Merck Sharp & Dohme, Novartis, Pfizer, progressive increase in the mean change in the Regeneron, Roche, Sanofi-Genzyme, Stiefel, Sun predicted DLQI scores among patients who Pharma, Takeda, UCB, and Valeant. Ahmed M. discontinued versus those who remained on Soliman is an employee of AbbVie, Inc. and treatment. The predicted HRQL deteriorated owns stocks in AbbVie, Inc. Nicolae Done, more rapidly than the PASI score, an objective Christopher Carley, and Esteban Lemus Wirtz measure of disease. are employees of Analysis Group, Inc., which received payment from AbbVie Inc. to conduct this research. Luis Puig received grants and personal fees from AbbVie during the conduct ACKNOWLEDGEMENTS of the study; grants and personal fees from Almirall, Amgen, Boehringer Ingelheim, Cel- gene, Janssen, LEO Pharma, Lilly, Novartis, Pfi- Funding. Sponsorship for this study and zer, Regeneron, Sanofi, Roche, and UCB outside Rapid Service Fee were funded by AbbVie Inc. the submitted work; and personal fees from The study sponsor contributed to the design; Sandoz, Merck-Serono, Merck Sharp & Dohme, participated in collection, analysis, and inter- Mylan, and Samsung-Bioepis outside the sub- pretation of data; and participated in writing, mitted work reviewing, and approval of the final version. Compliance with Ethics Guidelines. This Authorship. All named authors meet the article is based on previously conducted studies International Committee of Medical Journal and does not contain any new studies with Editors (ICMJE) criteria for authorship for this human participants performed by any of the article, take responsibility for the integrity of authors. The trials included in this study were the work as a whole, and have given their approved by each site’s institutional review approval for this version to be published. board and was performed in accordance with Authors’ contributions. KAP, AMS, ND, and the Helsinki Declaration of 1964, and its later LP contributed to study conceptualization and amendments. design. ND, CC, and ELW performed or sup- Prior presentation. Presented at the 2021 ported the formal analyses. All authors (KAP, American Academy of Dermatology Virtual AMS, ND, CC, ELW, and LP) contributed to the Meeting. data interpretation, critically reviewed the manuscript, and approved the submitted Data availability. AbbVie is committed to manuscript version. responsible data sharing regarding the clinical trials we sponsor. This includes access to anon- Medical writing, editorial, and other assis- ymized, individual and trial-level data (analysis tance. The authors would like to thank Gloria data sets), as well as other information (e.g., DeWalt from Analysis Group, Inc. for providing protocols and Clinical Study Reports), as long as editorial assistance. Support for this assistance the trials are not part of an ongoing or planned was provided by AbbVie Inc. regulatory submission. This includes requests Disclosures. Kim A. Papp received honoraria for clinical trial data for unlicensed products or fees for serving on advisory boards, or as a and indications. This clinical trial data can be speaker, or as a consultant, and grants as an requested by any qualified researchers who investigator from AbbVie, Amgen, Astellas, engage in rigorous, independent scientific Baxalta, Baxter, Boehringer Ingelheim, Bristol- research, and will be provided following review 1302 Dermatol Ther (Heidelb) (2021) 11:1291–1304 and approval of a research proposal and Statis- 4. Founta O, Adamzik K, Tobin A-M, Kirby B, Hevey D. tical Analysis Plan (SAP) and execution of a Data Psychological distress, alexithymia and alcohol misuse in patients with psoriasis: a cross-sectional Sharing Agreement (DSA). Data requests can be study. J Clin Psychol Med Settings. 2019;26(2): submitted at any time and the data will be 200–19. accessible for 12 months, with possible exten- sions considered. For more information on the 5. Yeung H, Takeshita J, Mehta NN, Kimmel SE, Ogdie A, Margolis DJ, et al. Psoriasis severity and the process, or to submit a request, visit the fol- prevalence of major medical comorbidity: a popu- lowing link: https://www.abbvie.com/our- lation-based study. JAMA Dermatol. 2013;149(10): science/clinical-trials/clinical-trials-data-and- 1173–9. information-sharing/data-and-information- 6. Koo J. Population-based epidemiologic study of sharing-with-qualified-researchers.html. psoriasis with emphasis on quality of life assess- ment. Dermatol Clin. 1996;14(3):485–96. Open Access. This article is licensed under a Creative Commons Attribution-NonCommer- 7. Michalek IM, Loring B, John SM. A systematic review of worldwide epidemiology of psoriasis. J Eur cial 4.0 International License, which permits Acad Dermatol Venereol. 2017;31(2):205–12. any non-commercial use, sharing, adaptation, distribution and reproduction in any medium 8. Singh S, Young P, Armstrong AW. An update on or format, as long as you give appropriate credit psoriasis and metabolic syndrome: a meta-analysis of observational studies. PLoS One. 2017;12(7): to the original author(s) and the source, provide e0181039. a link to the Creative Commons licence, and indicate if changes were made. The images or 9. Gisondi P, Fostini AC, Fossa I, Girolomoni G, Tar- other third party material in this article are gher G. Psoriasis and the metabolic syndrome. Clin Dermatol. 2018;36(1):21–8. included in the article’s Creative Commons licence, unless indicated otherwise in a credit 10. Levine D, Gottlieb A. Evaluation and management line to the material. If material is not included of psoriasis: an internist’s guide. Med Clin North in the article’s Creative Commons licence and Am. 2009;93(6):1291–303. your intended use is not permitted by statutory 11. Svoboda SA, Ghamrawi RI, Owusu DA, Feldman SR. regulation or exceeds the permitted use, you Treatment goals in psoriasis: which outcomes will need to obtain permission directly from the matter most? Am J Clin Dermatol. 2020;21(4): copyright holder. To view a copy of this licence, 505–11. visit http://creativecommons.org/licenses/by- 12. Fredriksson T, Pettersson U. Severe psoriasis—oral nc/4.0/. therapy with a new retinoid. Dermatology. 1978;157(4):238–44. 13. CfMPfHUC EMA. Guideline on clinical investiga- REFERENCES tion of medicinal products indicated or the treat- ment of psoriasis. London: EMEA; 2004. 1. Conrad C, Gilliet M. Psoriasis: from pathogenesis to 14. Lebwohl MG, Bachelez H, Barker J, Girolomoni G, targeted therapies. Clin Rev Allergy Immunol. Kavanaugh A, Langley RG, et al. Patient perspec- 2018;54(1):102–13. tives in the management of psoriasis: results from the population-based Multinational Assessment of 2. Donigan JM, Pascoe VL, Kimball AB. Psoriasis and Psoriasis and Psoriatic Arthritis Survey. J Am Acad herpes simplex virus are highly stigmatizing com- Dermatol. 2014;70(5):871–81-e1–30. pared with other common dermatologic condi- tions: a survey-based study. J Am Acad Dermatol. 15. Takeshita J, Callis Duffin K, Shin DB, Krueger GG, 2015;73(3):525–6. Robertson AD, Troxel AB, et al. Patient-reported outcomes for psoriasis patients with clear versus 3. de Korte J, Mombers FMC, Bos JD, Sprangers MAG. almost clear skin in the clinical setting. J Am Acad Quality of life in patients with psoriasis: a system- Dermatol. 2014;71(4):633–41. atic literature review. J Investig Dermatol Symp Proc. 2004;9(2):140–7. 16. Finlay AY, Khan G. Dermatology Life Quality Index (DLQI)—a simple practical measure for routine clinical use. Clin Exp Dermatol. 1994;19(3):210–6. Dermatol Ther (Heidelb) (2021) 11:1291–1304 1303 17. Li W, Ghamrawi R, Haidari W, Feldman SR. Risan- 26. Doshi JA, Takeshita J, Pinto L, Li P, Yu X, Rao P, kizumab for the treatment of moderate to severe et al. Biologic therapy adherence, discontinuation, plaque psoriasis. Ann Pharmacother. 2020;54(4): switching, and restarting among patients with 380–7. psoriasis in the US Medicare population. J Am Acad Dermatol. 2016;74(6):1057-65 e4. 18. Blauvelt A, Leonardi CL, Gooderham M, Papp KA, Philipp S, Wu JJ, et al. Efficacy and safety of con- 27. Bonafede M, Johnson BH, Fox KM, Watson C, tinuous risankizumab therapy vs treatment with- Gandra SR. Treatment patterns with etanercept and drawal in patients with moderate to severe plaque adalimumab for psoriatic diseases in a real-world psoriasis: a phase 3 randomized clinical trial. JAMA setting. J Dermatol Treat. 2013;24(5):369–73. Dermatol. 2020;156(6):649–58. 28. Yeung H, Wan J, Van Voorhees AS, Callis Duffin K, 19. Gordon KB, Strober B, Lebwohl M, Augustin M, Krueger GG, Kalb RE, et al. Patient-reported reasons Blauvelt A, Poulin Y, et al. Efficacy and safety of for the discontinuation of commonly used treat- risankizumab in moderate-to-severe plaque psoria- ments for moderate to severe psoriasis. J Am Acad sis (UltIMMa-1 and UltIMMa-2): results from two Dermatol. 2013;68(1):64–72. double-blind, randomised, placebo-controlled and ustekinumab-controlled phase 3 trials. Lancet. 29. Belinchon I, Ramos JM, Carretero G, Ferrandiz C, 2018;392(10148):650–61. Rivera R, Dauden E, et al. Adverse events associated with discontinuation of the biologics/classic sys- 20. Reich K, Gooderham M, Thaçi D, Crowley JJ, Ryan temic treatments for moderate-to-severe plaque C, Krueger JG, et al. Risankizumab compared with psoriasis: data from the Spanish Biologics Registry. adalimumab in patients with moderate-to-severe Biobadaderm J Eur Acad Dermatol Venereol. plaque psoriasis (IMMvent): a randomised, double- 2017;31(10):1700–8. blind, active-comparator-controlled phase 3 trial. Lancet. 2019;394(10198):576–86. 30. Poulin Y, Sheth P, Gu Y, Teixeira HD. Health-re- lated quality of life worsens disproportionately to 21. Risankizumab meets all primary endpoints report- objective signs of psoriasis after withdrawal of ing positive results in fourth pivotal phase 3 psori- adalimumab therapy. Dermatol Ther (Heidelb). asis study. Available at: https://news.abbvie.com/ 2014;4(1):33–42. news/risankizumab-meets-all-primary-endpoints- reporting-positive-results-in-fourth-pivotal-phase- 31. Ghajarzadeh M, Ghiasi M, Kheirkhah S. Associa- 3-psoriasis-study.htm [press release]. Accessed date tions between skin diseases and quality of life: a 4 Dec 2017 comparison of psoriasis, vitiligo, and alopecia areata. Acta Med Iran. 2012;1:511–5. 22. Viswanathan HN, Chau D, Milmont CE, Yang W, Erondu N, Revicki DA, et al. Total skin clearance 32. Daudén E, Pujol RM, Sánchez-Carazo JL, Toribio J, results in improvements in health-related quality of Vanaclocha F, Puig L, et al. Demographic charac- life and reduced symptom severity among patients teristics and health-related quality of life of patients with moderate to severe psoriasis. J Dermatol Treat. with moderate-to-severe psoriasis: the VACAP 2015;26(3):235–9. study. Actas Dermosifiliogr. 2013;104(9):807–14. 23. Hsu DY, Gniadecki R. Patient adherence to biologic 33. Herédi E, Rencz F, Balogh O, Gulácsi L, Herszényi K, agents in psoriasis. Dermatology. 2016;232(3): Holló P, et al. Exploring the relationship between 326–33. EQ-5D, DLQI and PASI, and mapping EQ-5D utili- ties: a cross-sectional study in psoriasis from Hun- 24. Mrowietz U, De Jong E, Kragballe K, Langley R, Nast gary. Eur J Health Econ. 2014;15(1):111–9. A, Puig L, et al. A consensus report on appropriate treatment optimization and transitioning in the 34. Palotai T, Szepetowski JC, Pec J, Arenberger P, management of moderate-to-severe plaque psoria- Giurcaneanu C, Gyulai R, et al. A survey of disease sis. J Eur Acad Dermatol Venereol. 2014;28(4): severity, quality of life, and treatment patterns of 438–53. biologically naive patients with psoriasis in central and eastern Europe. Acta Dermatovenerol Croat. 25. Daudén E, Puig L, Ferrándiz C, Sánchez-Carazo J, 2010;18(3):1. Hernanz-Hermosa J, Dermatology SPGotSAo, et al. Consensus document on the evaluation and treat- 35. Hawro T, Zalewska A, Hawro M, Kaszuba A, Kro- ment of moderate-to-severe psoriasis: Psoriasis likowska M, Maurer M. Impact of psoriasis severity Group of the Spanish Academy of Dermatology and on family income and quality of life. J Eur Acad Venereology. J Eur Acad Dermatol Venereol. Dermatol Venereol. 2015;29(3):438–43. 2016;30:1–18. 36. Moradi M, Rencz F, Brodszky V, Moradi A, Balogh O, Gulácsi L. Health status and quality of life in 1304 Dermatol Ther (Heidelb) (2021) 11:1291–1304 patients with psoriasis: an Iranian cross-sectional 41. Bronsard V, Paul C, Prey S, Puzenat E, Gourraud PA, survey. Arch Iran Med. 2015;18(3):1. Aractingi S, et al. What are the best outcome mea- sures for assessing quality of life in plaque type 37. Mabuchi T, Yamaoka H, Kojima T, Ikoma N, Aka- psoriasis? A systematic review of the literature. J Eur saka E, Ozawa A. Psoriasis affects patient’s quality of Acad Dermatol Venereol. 2010;24:17–22. life more seriously in female than in male in Japan. Tokai J Exp Clin Med. 2012;37(3):84–8. 42. Twiss J, Meads DM, Preston EP, Crawford SR, McKenna SP. Can we rely on the Dermatology Life 38. Tsai T-F, Ho J-C, Chen Y-J, Hsiao P-F, Lee W-R, Chi Quality Index as a measure of the impact of psori- C-C, et al. Health-related quality of life among asis or atopic dermatitis? J Invest Dermatol. patients with moderate-to-severe plaque psoriasis in 2012;132(1):76–84. Taiwan. Dermatol Sin. 2018;36(4):190–5. 43. Berth-Jones J, Grotzinger K, Rainville C, Pham B, 39. Robinson A, Kardos M, Kimball AB. Physician Glo- Huang J, Daly S, et al. A study examining inter- and bal Assessment (PGA) and Psoriasis Area and intrarater reliability of three scales for measuring Severity Index (PASI): why do both? A systematic severity of psoriasis: Psoriasis Area and Severity analysis of randomized controlled trials of biologic Index, Physician’s Global Assessment and Lattice agents for moderate to severe plaque psoriasis. J Am System Physician’s Global Assessment. Br J Derma- Acad Dermatol. 2012;66(3):369–75. tol. 2006;155(4):707–13. 40. Geale K, Henriksson M, Schmitt-Egenolf M. How is 44. Langley RG, Ellis CN. Evaluating psoriasis with disease severity associated with quality of life in Psoriasis Area and Severity Index, Psoriasis Global psoriasis patients? Evidence from a longitudinal Assessment, and Lattice System Physician’s Global population-based study in Sweden. Health Qual Assessment. J Am Acad Dermatol. 2004;51(4): Life Outcomes. 2017;15(1):151. 563–9.