Abstract
Introduction
Switching biologics within or across classes can improve outcomes for patients with psoriasis who failed to meet their treatment goals on their original therapy. The objective of this study was to identify real-world baseline features which are associated with switching psoriasis therapies following sustained use of a biologic therapy.
Methods
The study was a retrospective analysis of the prospective, multicenter, non-interventional PPD™ CorEvitas™ Psoriasis Registry cohort. Patient sociodemographics, comorbidities, treatment history, disease activity, and patient-reported outcome measures were assessed at baseline visits, along with changes in disease activity and treatment at follow-up visits. Patients were classified at each follow-up visit as either switchers from one biologic therapy to another or non-switchers. Three analytic strategies—logistic regression, random forest, and decision trees—were used to identify features associated with switching.
Results
Patients contributed 14,729 follow-up visits, of which 995 episodes (6.8%) reflected a switch in biologic therapy. In logistic regression models, statistically significant associations with switching were seen for features including body surface area (BSA) involvement at baseline, change in BSA involvement from baseline to follow-up, and addition of at least one non-biologic systemic medication to treatment between baseline and follow-up. In random forest estimations, these three variables along with patient-reported fatigue and quality of life were determined to be most important. Finally, in the decision tree analysis, four subgroups of patients with moderate/severe BSA involvement at baseline in combination with other specific variables were identified as having a > 50% likelihood of switching.
Conclusion
Identification and recognition of these features and combinations thereof can facilitate shared decision-making between clinicians and patients to improve both outcomes of and patient satisfaction with biologic therapy.
Supplementary Information
The online version contains supplementary material available at 10.1007/s13555-025-01646-1.
Keywords: Biologic therapy, Decision tree analysis, Psoriasis, Random forest analysis, Therapy switching
Plain Language Summary
Switching biologic medications can help patients with psoriasis who are not achieving their treatment goals with their current therapy after sustained (at least 6 months) use of their current therapy. This study aimed to identify common characteristics of patients from the PPD™ CorEvitas™ Psoriasis Registry who switched their psoriasis treatments after using a biologic therapy for some time; that is, to identify factors, other than waning effectiveness, that influence therapy switch after some success of their given therapy. Researchers looked at patient demographics, other health conditions, treatment history, disease activity, and patient-reported outcomes at the start of the study and during follow-up visits. Patients were categorized as either switchers (those who changed biologic therapies) or non-switchers at each follow-up visit. Three statistical approaches—logistic regression, random forest analysis, and decision tree analysis—were used to find factors linked to switching treatments. Logistic regression and random forest analyses highlighted factors such as the extent of skin involvement (body surface area, BSA) at the start, changes in BSA over time, adding non-biologic systemic medications, patient-reported fatigue and quality of life as being significantly associated with switching. Decision tree analyses identified four distinct patient groups with moderate/severe BSA involvement at baseline and other specific factors as having a higher likelihood of switching. Recognizing these characteristics can help doctors and patients with shared decision-making to meet patient therapy needs.
Supplementary Information
The online version contains supplementary material available at 10.1007/s13555-025-01646-1.
Key Summary Points
| Why carry out this study? |
| Failure to switch therapies when medically appropriate is a potential unmet need among patients with psoriasis. |
| A registry design was used to identify real-world baseline patient and disease characteristics associated with switching psoriasis therapies. |
| What was learned from the study? |
| Significant characteristics, commonly found in electronic health records, included greater body surface area percentage and worse patient-reported outcomes at baseline, and the addition of a concomitant non-biologic systemic medication between baseline and follow-up. |
| Clinicians can use the identified individual characteristics and combinations thereof to inform shared decision-making on treatment pathways. |
Introduction
Psoriasis is a chronic inflammatory disease characterized by skin lesions that can affect any area of the body [1, 2]. Treatment for psoriasis is generally based on disease activity, such that patients with moderate-to-severe disease may require advanced therapies and frequent therapy switches to find an effective treatment [1, 3]. These therapies may be biologic drugs, which inhibit specific inflammatory mediators of psoriasis. Alternatively, non-biologic systemic medications (NBSM) such as small molecule drugs may be considered instead of or in addition to biologics [4]. Treatment decisions also can be influenced by the area affected by psoriasis and the impact of the disease on patients’ lives [1].
The National Psoriasis Foundation (NPF) established psoriasis treatment goals for the USA [5], with the target response defined as reducing a patient’s psoriasis to ≤ 1% body surface area (BSA) involvement within 3 months after starting a new treatment. If an acceptable response (BSA involvement ≤ 3% or 75% improvement in BSA involvement) has not been met by 3 months, or the target response has not been met by 6 months, it is recommended that the provider change the dose, add a new treatment, or switch to a different treatment. Evaluation should continue every 6 months to determine whether the patient is maintaining the target response (≤ 1% BSA involvement). A number of observational studies have shown that switching biologics within or across classes can improve outcomes for patients who failed to meet their treatment goals on their original therapy [6].
However, failure to switch therapies when medically appropriate is a potential unmet need among patients with psoriasis [7]. For patients with psoriasis with an inadequate response within their first year of starting a biologic treatment, the median time to switch was approximately 5.8 months [8]. Inadequate response was not defined using clinical disease activity measures but by diagnosis codes on insurance claims [8]. Low adherence (proportion of days covered < 80%), newly added concomitant therapy, and increased dose have served as proxies for lack of skin clearance [9]. Much of the current research has focused on the economic burden of switching therapies rather than the burden on the patient and healthcare system of delaying the switch [10–12].
This study used a registry-based design to directly observe clinical measures, patient-reported outcomes, and treatments, filling a gap in the current literature. The primary objective of this study was to identify drug-agnostic baseline features available to clinicians in a real-world setting, such as patient characteristics and disease activity measures, which are associated with switching therapies following sustained use of a biologic therapy.
Methods
Data Source
The PPD™ CorEvitas™ Psoriasis Registry is a prospective, multicenter, non-interventional registry of patients with psoriasis under dermatologist care [13]. Enrolled patients were ≥ 18 years old, had psoriasis diagnosed by a dermatologist, and initiated or switched to an approved systemic or biologic psoriasis treatment as prescribed by the dermatologist at registry enrollment or within 12 months before enrollment. Data were collected from dermatologists and patients during routine clinical visits occurring at approximately 6-month intervals.
Study Design and Population
Patients included in the analyses resided in the USA, were on an approved biologic medication for at least 6 months of continuous use at the time of a registry visit (baseline visit), and had at least one follow-up visit approximately 6 months (range 5–9 months) after the baseline visit between July 1, 2019 and August 31, 2023. During the study period, each eligible pair of baseline and follow-up visits for a patient was considered an exposure episode (the unit of analysis) regardless of whether a switch occurred. All eligible exposure episodes were included. Patients may have contributed multiple independent exposure episodes if there were multiple baseline/follow-up visit pairs that met the inclusion criteria, including BSA and Psoriasis Area and Severity Index (PASI) values at the baseline and follow-up visits (Supplemental Figs. S1–S3).
Outcomes
The primary outcome was an indicator of whether a patient switched from one biologic therapy to another during the course of regular clinical treatment between baseline and follow-up. Patients were classified as either switchers or non-switchers at their follow-up visit for each contributed exposure episode.
Patients were considered a switcher for a given exposure episode if they (1) discontinued their biologic therapy at or before the 6-month follow-up visit and initiated or restarted an alternative biologic or small molecule therapy or (2) added a second biologic or small molecule medication, without an observed discontinuation of the original biologic therapy. Switches to or from a biosimilar and/or the original biologic medication were considered a continuation of the same therapy. Patients who discontinued therapy at or before the 6-month follow-up and did not start a new therapy were classified as non-switchers.
Covariates
Baseline visit variables included sociodemographic and lifestyle characteristics (e.g., age), comorbidities [e.g., psoriatic arthritis (PsA)], psoriasis disease activity measures (e.g., BSA involvement), past and concomitant psoriasis therapies (e.g., topicals), and patient-reported outcome measures [e.g., Dermatology Life Quality Index [DLQI] and Patient Global Assessment (PGA)]. Disease activity measures were also assessed at the follow-up visit. BSA involvement was categorized into mild (< 3%), moderate (3–10%), or severe (> 10%), and change in BSA category from baseline to follow-up was assessed. See Supplemental Table S1 for additional variable details.
Statistical Analyses
Patient characteristics at the baseline visit and disease activity measures at follow-up were summarized using descriptive statistics, both overall and stratified by switching status. Three analytic strategies were used to identify features associated with switching.
First, multivariable logistic regression was used to quantify magnitude and direction of associations observed between the features and switching. Model-estimated odds ratios (ORs), 95% confidence intervals (CIs), and P values with a significance level of 0.05 were calculated for all variables under consideration. Variable selection using a Bayesian information criterion with stepwise elimination was performed to determine the variables associated with switching. The relative importance of each variable in the logistic regression model was assessed by the absolute value of the z statistic. After the primary model was identified, two sensitivity analyses were conducted: (1) all two-way interaction effects were considered for the logistic regression model; and (2) patients in the study cohort were linked to commercial administrative claims to include additional healthcare utilization measures.
The second strategy, estimation of a random forest, was used to partition the patient variables from the logistic regression model and test importance. In the random forest algorithm, a collection of non-linear prediction trees divided the variables into optimal splits until partitions were identified that selected the most homogenous sub-nodes [14–18]. Next, this process was iterated to grow many prediction trees to create the “forest”. Once the algorithm was trained, variable importance of the resulting forest was calculated by identifying increased prediction errors when the data for each individual variable was permuted while all other variables were left fixed. Accumulated local effects (ALE) and SHapley Additive exPlanations (SHAP) plots were produced for interpretability.
Finally, a decision tree algorithm was used to identify combinations of features (using variables identified by the logistic regression model described above) that distinguished patients likely to switch biologic therapy. Decision trees, much like random forests, are binary recursive partitioning methods that help detect complex interaction effects. They avoid bias selection toward variables with many possible splits by using a permutation test to decide which variables to split and recurse [19–21].
R Version 4.3.2 (The R Foundation for Statistical Computing, Vienna, Austria) and SAS 9.4 (SAS Institute, Inc, Cary, NC, USA) were used for the analyses.
Ethics
The study was performed in accordance with the Declaration of Helsinki and the Guidelines for Good Pharmacoepidemiology Practice (GPP). All participating investigators were required to obtain full board approval for conducting non-interventional research involving human subjects with a limited dataset. Sponsor approval and continuing review was obtained through a central institutional review board (IRB, Advarra, Protocol number Pro00051221). For academic investigative sites that did not receive a waiver to use the central IRB, full board approval was obtained from the respective governing IRBs and documentation of approval was submitted to Thermo Fisher Scientific prior to the initiation of any study procedures. All patients in the registry were required to provide written informed consent and authorization prior to participating.
Results
Baseline Characteristics
Patients (N = 5948) contributed 14,729 exposure episodes that met the inclusion criteria. Of these, 995 episodes (6.8%) had a switch in their biologic therapy, while the remaining 13,734 episodes (93.2%) were non-switchers. There were no significant differences (standardized differences [std diff] < 0.1) in demographics between switchers and non-switchers (Table 1, Supplemental Table S2). Baseline variables with significant differences included BSA involvement, number of prior medications including NBSM and topicals, and many patient-reported outcome measures (Table 1, Supplemental Table S2). Further, there was a statistically significant difference in change in BSA involvement from baseline to follow-up between switchers and non-switchers (std diff = 0.180), with 24.5% of switchers having an increase in BSA category, indicating worsening BSA involvement, while only 7.4% of non-switchers had a similar shift in category towards worsening BSA involvement (Supplemental Table S3).
Table 1.
Selected patient characteristics at baseline, overall and stratified by switching status
| Variable | Overall N = 14,729 |
Non-switchers N = 13,734 |
Switchers N = 995 |
Standardized difference |
|---|---|---|---|---|
| Demographic characteristics | ||||
| Age, mean (SD) | 53.2 (14.3) | 53.3 (14.4) | 52.0 (13.7) | 0.091 |
| Median (Q1, Q3) | 54.0 (43.0, 64.0) | 54.0 (43.0, 64.0) | 53.0 (43.0, 62.0) | |
| Female, n (%) | 6714 (45.6) | 6196 (45.1) | 518 (52.1) | 0.035 |
| Race, n (%a) | 0.030 | |||
| White | 11,648 (79.2) | 10,894 (79.4) | 754 (75.9) | |
| Black | 543 (3.7) | 489 (3.6) | 54 (5.4) | |
| Asian | 1485 (10.1) | 1382 (10.1) | 103 (10.4) | |
| Other | 1031 (7.) | 948 (6.9) | 83 (8.4) | |
| Ethnicity (Hispanic), n (%) | 1521 (10.5) | 1419 (10.5) | 102 (10.4) | 0.001 |
| Employment (full time), n (%) | 8645 (58.9) | 8070 (59.0) | 575 (58.1) | 0.005 |
| Year of baseline visit, n (%) | 0.029 | |||
| 2019 | 1848 (12.5) | 1689 (12.3) | 159 (16.0) | |
| 2020 | 4231 (28.7) | 3944 (28.7) | 287 (28.8) | |
| 2021 | 4571 (31.0) | 4277 (31.1) | 294 (29.5) | |
| 2022 | 4079 (27.7) | 3824 (27.8) | 255 (25.6) | |
| Baseline medication class, n (%) | 0.078 | |||
| TNFi | 1770 (12.0) | 1602 (11.7) | 168 (16.9) | |
| IL-17i | 5414 (36.8) | 4959 (36.1) | 445 (44.7) | |
| IL-23i | 5797 (39.4) | 5541 (40.3) | 256 (25.7) | |
| Other | 1758 (11.9) | 1632 (11.9) | 126 (12.7) | |
| Disease characteristics | ||||
| Psoriatic arthritis, n (%) | 6274 (42.9) | 5758 (42.2) | 516 (52.1) | 0.050 |
| BSA (% involvement), mean (SD) | 1.8 (4.4) | 1.6 (4.1) | 3.8 (7.1) | 0.498 |
| Median (Q1, Q3) | 1.0 (0.0, 2.0) | 1.0 (0.0, 1.8) | 2.0 (0.0, 5.0) | |
| Mild disease (< 3%), n (%) | 12,152 (82.5) | 11,542 (84.0) | 610 (61.3) | 0.155 |
| Moderate disease (3–10%), n (%) | 2225 (15.1) | 1,914 (13.9) | 311 (31.3) | |
| Severe disease (> 10%), n (%) | 352 (2.4) | 278 (2.0) | 74 (7.4) | |
| At least 1 prior biologic therapy, n (%) | 1815 (12.3) | 1647 (12.0) | 168 (16.9) | 0.024 |
| Duration of biologic therapy at baseline | 0.064 | |||
| 5 to < 12 months, n (%) | 5046 (34.3) | 4595 (33.5) | 451 (45.3) | |
| 12 to < 24 months, n (%) | 5045 (34.3) | 4740 (34.5) | 305 (30.7) | |
| 24+ months, n (%) | 4638 (31.5) | 4399 (32.0) | 239 (24.0) | |
| Number of prior NBSM, mean (SD) | 0.1 (0.4) | 0.1 (0.4) | 0.2 (0.5) | 0.148 |
| Median (Q1, Q3) | 0.0 (0.0, 0.0) | 0.0 (0.0, 0.0) | 0.0 (0.0, 0.0) | |
| NBSM addedb, n (%) | 83 (0.6) | 56 (0.4) | 27 (2.7) | 0.077 |
| Number of prior topical therapies, mean (SD) | 1.2 (2.0) | 1.2 (1.9) | 1.6 (2.2) | 0.179 |
| Median (Q1, Q3) | 0.0 (0.0, 3.0) | 0.0 (0.0, 3.0) | 0.0 (0.0, 3.0) | |
| Patient-reported outcomes | ||||
| Patient global assessment, mean (SD) | 15.1 (22.7) | 14.3 (22.2) | 25.9 (26.5) | 0.517 |
| Median (Q1, Q3) | 5.0 (0.0, 20.0) | 5.0 (0.0, 15.0) | 15.0 (5.0, 40.0) | |
| DLQI, mean (SD) | 1.9 (3.2) | 1.7 (3.0) | 3.5 (4.5) | 0.560 |
| Median (Q1, Q3) | 1.0 (0.0, 2.0) | 1.0 (0.0, 2.0) | 2.0 (1.0, 5.0) | |
| Fatigue, mean (SD) | 18.6 (24.2) | 18.0 (23.8) | 27.0 (27.2) | 0.372 |
| Median (Q1, Q3) | 7.0 (0.0, 30.0) | 5.0 (0.0, 25.0) | 20.0 (4.0, 50.0) | |
| Itch/pruritus, mean (SD) | 13.6 (22.2) | 12.8 (21.5) | 24.5 (28.5) | 0.531 |
| Median (Q1, Q3) | 5.0 (0.0, 15.0) | 5.0 (0.0, 15.0) | 10.0 (2.8, 40.0) | |
| EQ-5D, mean (SD) | 80.2 (17.9) | 80.6 (17.8) | 75.1 (18.9) | 0.306 |
| Median (Q1, Q3) | 85.0 (75.0, 90.0) | 85.0 (75.0, 91.0) | 80.0 (70.0, 90.0) | |
BSA body surface area, DLQI Dermatology Life Quality Index, EQ-5D EuroQOL Five Dimensions Questionnaire, IL-17i interleukin-17 inhibitor, IL-23i interleukin-23 inhibitor, NBSM non-biologic systemic medication, Q1 1st quartile, Q3 3rd quartile, SD standard deviation, TNFi tumor necrosis factor inhibitor
aPercentages are among those with non-missing values. Percentages may not sum to 100 due to rounding
bMeasures from baseline to follow-up do not include the status at the endpoints. That is, they measure changes between the 2 registry visits, not including those occurring at the given visits
Logistic Regression
Statistically significant associations with switching were seen for BSA involvement at baseline (moderate disease: OR 4.19 [95% CI 3.40, 5.17]; severe disease: OR 5.94 [95% CI 4.19, 8.41]), change in BSA involvement from baseline to follow-up (decrease in category: OR 0.69 [95% CI 0.55, 0.88]; increase in category: OR 6.23 [95% CI 5.17, 7.49]), and addition of at least one NBSM to the therapy regimen between baseline and follow-up (OR 4.87 [95% CI 2.92, 8.14]) (Table 2). Variable importance confirmed these key relationships (Supplemental Fig. S4).
Table 2.
Logistic regression model regressing switch status on patient characteristics, clinical features, and treatment history (final reduced model)
| Variable | OR | 95% CI | P value |
|---|---|---|---|
| Change in BSA involvement (ref: No change) | |||
| Decrease in BSA involvement | 0.69 | 0.55, 0.88 | 0.002 |
| Increase in BSA involvement | 6.23 | 5.17, 7.49 | < 0.001 |
| BSA category (ref: Mild [< 3%]) | |||
| Moderate (3–10%) | 4.19 | 3.40, 5.17 | < 0.001 |
| Severe (> 10%) | 5.94 | 4.19, 8.41 | < 0.001 |
| Patient global assessment | 1.01 | 1.00, 1.01 | < 0.001 |
| NBSM added | 4.87 | 2.92, 8.14 | < 0.001 |
| DLQI | 1.04 | 1.02, 1.06 | < 0.001 |
| Female | 1.29 | 1.12, 1.49 | < 0.001 |
| Age | 0.99 | 0.99, 1.00 | 0.001 |
| Psoriatic arthritis | 1.35 | 1.16, 1.56 | < 0.001 |
| Duration of biologic therapy at baseline (Ref: 5 to < 12 months) | |||
| 12 to < 24 months | 0.75 | 0.64, 0.88 | 0.001 |
| 24+ months | 0.79 | 0.66, 0.94 | 0.008 |
| Ethnicity (Hispanic) | 0.72 | 0.56, 0.92 | 0.008 |
| Number of prior topicals | 1.05 | 1.02, 1.09 | 0.002 |
| Year of baseline visit (ref: 2019) | |||
| 2020 | 0.73 | 0.59, 0.92 | 0.006 |
| 2021 | 0.73 | 0.58, 0.91 | 0.004 |
| 2022 | 0.72 | 0.58, 0.91 | 0.005 |
| Employment (full time) | 1.18 | 1.01, 1.38 | 0.041 |
| EQ-5D | 1.00 | 0.99, 1.00 | 0.140 |
| Fatigue | 1.00 | 1.00, 1.01 | 0.146 |
BSA body surface area, CI confidence interval, EQ-5D EuroQOL Five Dimensions Questionnaire, DLQI Dermatology Life Quality Index, NBSM non-biologic systemic medication, OR odds ratio, ref reference level
Random Forest
Five variables were determined to be most important based on mean decreases in accuracy from the random forest estimations: (1) change in BSA category from baseline to follow-up; (2) BSA category; (3) fatigue; (4) addition of a NBSM between baseline and follow-up; and (5) DLQI (Table 3).
Table 3.
Variable importance for random forest model for prediction of biologic therapy switch at follow-up (final reduced model)
| Variable | Mean decrease accuracy | Mean decrease Ginia |
|---|---|---|
| Change in BSA involvement from baseline to follow-up | 38.1 | 55.9 |
| BSA involvement at baseline | 28.1 | 40.9 |
| Fatigue at baseline | 16.8 | 133.3 |
| NBSM added from baseline to follow-up | 16.2 | 16.1 |
| DLQI at baseline | 13.7 | 115.3 |
| Patient global assessment at baseline | 12.1 | 139.0 |
| Age at baseline | 12.1 | 181.6 |
| Employment at baseline | 11.4 | 29.7 |
| EQ-5D at baseline | 11.1 | 134.1 |
| Year of baseline visit | 3.6 | 78.8 |
| Ethnicity (Hispanic) at baseline | 2.7 | 18.2 |
| PsA at baseline | 2.6 | 29.4 |
| Number of prior topical therapies as assessed at baseline | 2.4 | 65.7 |
| Duration of biologic therapy at baseline | 1.3 | 50.7 |
| Gender at baseline | − 1.4 | 30.3 |
BSA body surface area, DLQI Dermatology Life Quality Index, EQ-5D EuroQOL Five Dimensions Questionnaire, NBSM non-biologic systemic medication, PsA psoriatic arthritis
aMean decrease in Gini impurity, a method of evaluating prediction fit, measures how much each variable contributes to the homogeneity of the nodes and leaves of a random forest. The smaller the Gini, the more “pure” the resulting forest
ALE plots demonstrated positive estimates, and therefore greater likelihood of switching, among exposure episodes with an increase or decrease in BSA involvement, moderate or severe BSA involvement at baseline, and addition of NBSM between baseline and follow-up. Negative estimates were seen for mild disease at baseline and no added NBSM between baseline and follow-up, indicating lower likelihood of switching. For fatigue and DLQI at baseline, lower values corresponded to a prediction less than the average feature effect (< 0) with a rapid increase as the measure worsened. Therefore, the relationship between each and switching was stronger for larger values of baseline fatigue and DLQI (Supplemental Fig. S5). Higher values of change in BSA involvement, BSA involvement at baseline, PGA, DLQI, and fatigue had positive SHAP values (Supplemental Fig. S6), indicating that they were most predictive of switching.
Decision Tree
Four subgroups with moderate/severe BSA involvement at baseline were identified in the decision tree analysis as having a very high likelihood of switching (> 50% switching). These were patients with (1) PsA, PGA < 20, DLQI ≥ 3, and aged 53–55 (71% switched); (2) PsA, PGA 14–19, DLQI between 2 and 3, and male (67% switched); (3) PGA ≥ 20, DLQI ≥ 13, ≥ 5 prior topicals, and no NBSM added between baseline and follow-up (67% switched); and (4) PGA ≥ 20, and added an NBSM between baseline and follow-up (56% switched) (Fig. 1).
Fig. 1.
Decision tree for patients with moderate to severe psoriasis disease. DLQI Dermatology Life Quality Index, NBSM non-biologic systemic medication, PsA psoriatic arthritis, Pt patient
Among those with mild BSA involvement at baseline, there were two subgroups of patients in which more than 50% switched biologic therapy at follow-up; those with (1) PGA < 8, NBSM added, and at least one prior topical (56% switched) and (2) PGA ≥ 8, DLQI 12 to < 15, aged < 71, fatigue < 61, and on their current therapy between 6 and 12 months at baseline (58% switched) (Supplemental Fig. S7).
Sensitivity Analyses
The primary logistic regression model was estimated with the addition of all possible two-way interactions (Supplemental Table S4). The statistically significant interaction terms identified within the final reduced model, along with the proportion of exposure episodes who switched therapies within each level, are presented (Table 4).
Table 4.
Subgroup summaries for levels associated with statistically significant interaction terms in the final reduced model
| Number | Overall population (%) | Switchers (%) | |
|---|---|---|---|
| Fatigue at baselinea × DLQI at baselinea | |||
| Fatigue ≤ 18.6, DLQI ≤ 1.85 | 6934 | 50.3 | 3.8 |
| Fatigue ≤ 18.6, DLQI > 1.85 | 1979 | 14.4 | 9.8 |
| Fatigue > 18.6, DLQI ≤ 1.85 | 2473 | 18.0 | 6.6 |
| Fatigue > 18.6, DLQI > 1.85 | 2386 | 17.3 | 12.8 |
| NBSM added from baseline to follow-up × age at baselinea | |||
| NBSM added, age ≤ 53.2 | 40 | 0.3 | 42.5 |
| NBSM added, age > 53.2 | 41 | 0.3 | 22.0 |
| No NBSM added, age ≤ 53.2 | 6639 | 48.2 | 6.9 |
| No NBSM added, age > 53.2 | 7052 | 51.2 | 6.3 |
| PsA at baseline × age at baselinea | |||
| PsA, age ≤ 53.2 | 2593 | 18.8 | 9.4 |
| PsA, age > 53.2 | 3353 | 24.3 | 7.4 |
| No PsA, age ≤ 53.2 | 4086 | 29.7 | 5.7 |
| No PsA, age > 53.2 | 3740 | 27.2 | 5.4 |
| Gender at baseline × duration of biologic therapy at baseline | |||
| Male, 5 to < 12 months | 2421 | 17.6 | 7.7 |
| Male, 12 to < 24 months | 2534 | 18.4 | 5.9 |
| Male, ≥ 24 months | 2509 | 18.2 | 4.1 |
| Female, 5 to < 12 months | 2276 | 16.5 | 10.0 |
| Female, 12 to < 24 months | 2145 | 15.6 | 6.2 |
| Female, ≥ 24 months | 1887 | 13.7 | 6.7 |
| Gender at baseline × fatigue at baselinea | |||
| Male, fatigue ≤ 18.6 | 5234 | 38.0 | 5.1 |
| Male, fatigue > 18.6 | 2230 | 16.2 | 7.7 |
| Female, fatigue ≤ 18.6 | 3679 | 26.7 | 5.1 |
| Female, fatigue > 18.6 | 2629 | 19.1 | 11.4 |
| Employment at baseline × fatigue at baselinea | |||
| Full time, fatigue ≤ 18.6 | 5462 | 39.7 | 5.0 |
| Full time, fatigue > 18.6 | 2710 | 19.7 | 10.0 |
| Not full time, fatigue ≤ 18.6 | 3451 | 25.1 | 5.2 |
| Not full time, fatigue > 18.6 | 2149 | 15.6 | 9.3 |
| PsA at baseline × PGA at baselinea | |||
| PsA, PGA ≤ 15.1 | 4054 | 29.4 | 6.0 |
| PsA, PGA > 15.1 | 1892 | 13.7 | 13.0 |
| No PsA, PGA ≤ 15.1 | 6165 | 44.8 | 3.7 |
| No PsA, PGA > 15.1 | 1661 | 12.1 | 12.2 |
DLQI Dermatology Life Quality Index, NBSM non-biologic systemic medication, PGA patient global assessment, PsA psoriatic arthritis
aFor the continuous variables, the cutoff associated with each is equivalent to the overall mean in the population
A small proportion of the exposure episodes (n = 1627, 11.0%) were linked to administrative claims from commercial insurance plans. Of these, 84 (5.2%) were switchers and the remaining were non-switchers (Supplemental Table S5). None of the healthcare utilization measures were statistically significant in the full logistic regression models (Supplemental Table S6).
Discussion
This study leveraged three different analytic approaches to quantifying which patient features were associated with a switch in biologic medications for patients who have been on their therapy long-term: logistic regression, random forests, and decision trees. Four features consistently identified as being associated with switching were higher disease activity as measured by BSA involvement at baseline, worse patient-reported outcomes (specifically non-zero PGA and DLQI) at baseline, and the addition of a concomitant NBSM between baseline and follow-up. Consistent with NPF treatment goals is the fact that BSA-related variables (at baseline and change from baseline to follow-up) were identified as important features associated with therapy switch [5]. These findings provide support for the NPF’s strategic plan goals to cure psoriatic disease entirely and optimize current health for everyone living with psoriatic disease [22].
The decision tree analysis identified rare but significant combinations of features subgroups most likely to switch therapies. Each of these subgroups was defined by moderate/severe BSA involvement, plus a combination of attributes similar to those that were significant in the logistic regression model and random forest algorithms. Similarly, the sensitivity analysis on the logistic regression analyses which identified interaction terms associated with switching also identified subsets of the population that were more likely to switch. These results may help clinicians in decision-making based on features discernible from standard electronic health record (EHR) information. For example, the importance of added NBSM in leading to switching is elucidated, with younger patients with NBSM added more likely to switch than older patients.
The measurement of switching and the features identified as important to it in this study differ from those in previous studies [8–12, 23–25]. Prior work generally examined the likelihood of any switch during a patient’s full follow-up period, describing that approximately 10–35% of patients switched medication at some point depending on length of follow-up [8, 10, 11, 23–25]. In contrast, this study examined the likelihood of switch at each individual exposure episode among the patients, following at least 6 months of therapy (6.8%). Additionally, patient features were derived from clinical and patient-reported outcomes data captured in a registry, rather than relying on treatment patterns derived from claims data as a proxy for clinical markers.
Strengths
The PPD CorEvitas Psoriasis Registry is a unique resource with large sample size and longitudinal follow-up on the real-world use of treatments for psoriasis. The registry enables examination of response patterns based on measures relevant to patient–physician encounters using clinical data that are not available in claims databases.
Strengths of this study include ease of interpretability for decision-makers. The intent of the study was to develop a simple drug-agnostic model, with features commonly found in EHRs, that can be used by healthcare providers to proactively identify leading indicators associated with biologic therapy switch. Therefore, overly complex models that are difficult to interpret and apply in the healthcare setting were avoided. In addition, the consistency of the findings across complementary analytic strategies suggested that the findings were robust.
Limitations
Limitations of the PPD CorEvitas Psoriasis Registry include generalizability; patients may not be representative of all adults with psoriasis. Additionally, the registry does not include measures of treatment adherence.
Limitations of the analysis include potential confounding by indication, which can occur when factors that determine physicians’ selection of a particular treatment or change in treatment are unobserved. It is also possible that drug benefits, design, and formulary status changes affected utilization patterns and, in turn, the number of patients who switch medications. Exclusion of other features such as individual therapies, treating physician/center, and biologic/class of the individual therapies may also introduce some bias. Roughly half of the eligible exposure periods were excluded from the cohort because the patient did not have a follow-up visit within 5–9 months of the index visit; however, all patients had been on therapy for at least 6 months prior to their index visit, so potential bias due to patients discontinuing therapy has been mitigated. Further, while some drugs may work faster than others, the effectiveness should be reasonably interchangeable for those on a drug for at least 6 months at baseline as an indicator of ongoing, sustained effectiveness. Moreover, the influence of physician preference is likely to be less impactful on biologic therapy switch, given the inclusion criteria of at least 6-month persistence at baseline. Finally, variables measured after baseline visit (e.g., change in BSA involvement) may be biased by endogeneity with the outcome of biologic therapy switch; however, these were important indicators of the worsening condition of the patient.
Implications for Clinical Practice
The findings of this study have significant implications for clinical practice. Although practices systematically capture the switching-associated disease severity metrics in EHRs, they often lack guidance on how to optimally utilize this information to improve patient care and return on investment. This study represents a potential practical application for such data through feature and subgroup identification. Additionally, the inclusion of patient-reported measures can help inform waning effectiveness when other clinical indicators may be sparse and help patients express their experience to assist in shared decision-making while minimizing the data collection burden on providers. Further, the data can be used in various ways, including consideration of the individual features identified in the analysis, as well as the rare but significant combinations of features identified in the decision tree analysis. This approach ensures that clinicians, regardless of their resources and metrics available for patients, can be better informed about the influence of clinical and personal features leading to the need for therapy switches. By providing multiple looks at patient features associated with biologic therapy switch, a broad toolbox of individual features and more complex patient profiles can be used by clinicians and treatment decision-makers to inform their decisions on treatment pathways for patients when they have a variety of information available through EHRs.
Conclusion
For patients with psoriasis on long-term biologic therapy, it can be difficult to identify appropriate timing to switch therapy when effectiveness wanes. Clinical factors such as changes in BSA involvement over time, escalating therapy by adding NBSM to the regimen, or potentially underdocumented factors such as fatigue and quality of life can signal decreasing effectiveness. It is beneficial for clinicians to understand which clinical characteristics and patient-reported outcomes from patient EHRs can help to identify patients with psoriasis who are likely to switch biologic therapies. Identification and recognition of these features can facilitate shared decision-making between clinicians and patients to improve patient satisfaction and outcomes.
Supplementary Information
Below is the link to the electronic supplementary material.
Acknowledgments
Medical Writing/Editorial Assistance
The authors acknowledge Tiffany Park, PharmD, the UCB Publications Lead for US Medical Affairs Immunology, Smyrna, GA. Medical writing assistance was provided by Kelly Strutz, PhD, of Thermo Fisher Scientific and funded by UCB, Inc. Editorial assistance was provided by Tyan Aleshire of Thermo Fisher Scientific and Ann Quan and Amanda Adams of Costello Medical, all funded by UCB, Inc. The authors would like to thank all the investigators, their clinical staff, and patients who participate in the PPD CorEvitas Psoriasis Registry.
Author Contributions
Andrea M. Austin made substantial contributions to the conception and design of the work, analysis and interpretation of data for the work, drafting and critical revision of the work for important intellectual content. Scott C. Henderson made substantial contributions to the conception and design of the work, analysis and interpretation of data for the work, drafting and critical revision of the work for important intellectual content. Natasha C. Trujillo made substantial contributions to the conception of the work, interpretation of data for the work, and critical revision of the work for important intellectual content. Robert Low made substantial contributions to the conception of the work, interpretation of data for the work, and critical revision of the work for important intellectual content. Melissa Eliot made substantial contributions to the design of the work, analysis and interpretation of data for the work, and critical revision of the work for important intellectual content. Sandra I. Main made substantial contributions to the design of the work and critical revision of the work for important intellectual content. Heather J. Litman made substantial contributions to the acquisition and interpretation of data for the work, and critical revision of the work for important intellectual content. Omeed Nabavian made substantial contributions to the conception and design of the work, interpretation of data for the work, and critical revision of the work for important intellectual content.
Funding
This study was sponsored by the PPD™ clinical research business of Thermo Fisher Scientific, Waltham, MA, USA, and the analysis was funded by UCB, Smyrna, GA, USA. The journal’s Rapid Service Fee was funded by UCB. Access to study data was limited to PPD™ CorEvitas™ Clinical Registries and our statisticians completed all the analysis; all authors contributed to the interpretation of the results. CorEvitas has been supported through contracted subscriptions in the last 2 years by AbbVie, Amgen, Inc., Arena, Boehringer Ingelheim, Bristol Myers Squibb, Chugai, Eli Lilly and Company, Genentech, GSK, Janssen Pharmaceuticals, Inc., LEO Pharma, Novartis, Ortho Dermatologics, Pfizer, Inc., Sun Pharmaceutical Industries Ltd., and UCB. The PPD CorEvitas Psoriasis Registry was developed in collaboration with the National Psoriasis Foundation (NPF).
Declarations
Conflict of Interest
Andrea M. Austin and Heather J. Litman are employees and shareholders of Thermo Fisher Scientific. Scott C. Henderson and Melissa Eliot are employees of Thermo Fisher Scientific. Natasha C. Trujillo, Robert Low and Omeed Nabavian are employees and shareholders of UCB. At the time of the study, Sandra I. Main was an employee of Thermo Fisher Scientific.
Ethical Approval
The study was performed in accordance with the Declaration of Helsinki and the Guidelines for Good Pharmacoepidemiology Practice (GPP). All participating investigators were required to obtain full board approval for conducting non-interventional research involving human subjects with a limited dataset. Sponsor approval and continuing review was obtained through a central institutional review board (IRB, Advarra, Protocol Number Pro00051221). For academic investigative sites that did not receive a waiver to use the central IRB, full board approval was obtained from the respective governing IRBs and documentation of approval was submitted to Thermo Fisher Scientific prior to the initiation of any study procedures. All patients in the registry were required to provide written informed consent and authorization prior to participating.
Footnotes
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
References
- 1.Griffiths CEM, Armstrong AW, Gudjonsson JE, Barker JNWN. Psoriasis. Lancet. 2021;397(10281):1301–15. [DOI] [PubMed] [Google Scholar]
- 2.Global report on psoriasis. World Health Organization. Updated October 26, 2016. https://www.who.int/publications/i/item/9789241565189. Accessed 19 Feb 2025.
- 3.Rendon A, Schäkel K. Psoriasis pathogenesis and treatment. Int J Mol Sci. 2019;20(6):1475. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Lin CP, Merola JF, Wallace EB. Current and emerging biologic and small molecule systemic treatment options for psoriasis and psoriatic arthritis. Curr Opin Pharmacol. 2022;67:102292. [DOI] [PubMed] [Google Scholar]
- 5.Armstrong AW, Siegel MP, Bagel J, et al. From the Medical Board of the National Psoriasis Foundation: treatment targets for plaque psoriasis. J Am Acad Dermatol. 2017;76(2):290–8. [DOI] [PubMed] [Google Scholar]
- 6.Kerdel F, Zaiac M. An evolution in switching therapy for psoriasis patients who fail to meet treatment goals. Dermatol Ther. 2015;28(6):390–403. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Feldman SR, Goffe B, Rice G, et al. The challenge of managing psoriasis: unmet medical needs and stakeholder perspectives. Am Health Drug Benefits. 2016;9(9):504–13. [PMC free article] [PubMed] [Google Scholar]
- 8.Thai S, Zhuo J, Zhong Y, et al. Real-world treatment patterns and healthcare costs in patients with psoriasis taking systemic oral or biologic therapies. J Dermatol Treat. 2023;34(1):217608. [DOI] [PubMed] [Google Scholar]
- 9.Birt J, Grabner M, Isenberg K, et al. Inadequate response among psoriatic arthritis patients prescribed advanced therapy in a real-world US commercially insured population [abstract]. Arthritis Rheumatol. 2020;72(suppl 10):0898. [DOI] [PubMed]
- 10.Feldman SR, Tian H, Wang X, Germino R. Health care utilization and cost associated with biologic treatment patterns among patients with moderate to severe psoriasis: analyses from a large US claims database. J Manag Care Spec Pharm. 2019;25(4):479–88. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Wu JJ, Patel M, Li C, Mandava MR, Armstrong AW. Real-world switch rates of biologics and associated costs in patients with psoriasis [abstract]. J Am Acad Dermatol. 2023;89(3 suppl 1):AB81. [Google Scholar]
- 12.Mallya U, Qureschi A, Lahoz R. The economic burden of switching biologics in psoriasis: a real-world analysis in the US population [abstract]. J Am Acad Dermatol. 2014;70(5 suppl 1):AB191. [Google Scholar]
- 13.Strober B, Karki C, Mason M, et al. Characterization of disease burden, comorbidities, and treatment use in a large, US-based cohort: results from the Corrona Psoriasis Registry. J Am Acad Dermatol. 2018;78(2):323–32. [DOI] [PubMed] [Google Scholar]
- 14.Austin AM, Ramkumar N, Gladders B, et al. Using a cohort study of diabetes and peripheral artery disease to compare logistic regression and machine learning via random forest modeling. BMC Med Res Methodol. 2022;22(1):300. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Breiman L. Random forests. Mach Learn. 2001;45(1):5–32. [Google Scholar]
- 16.Kirasich K, Smith T, Sadler B. Random forest vs logistic regression: binary classification for heterogeneous datasets. SMU Data Sci Rev. 2018;1(3):9. [Google Scholar]
- 17.Biau G. Analysis of a random forests model. J Mach Learn Res. 2012;13:1063–95. [Google Scholar]
- 18.Denil M, Matheson D, de Freitas N. Narrowing the gap: random forests in theory and in practice. PMLR. 2014;32(1):665–73. [Google Scholar]
- 19.Iancu A, Leb I, Prokosch HU, Rödle W. Machine learning in medication prescription: a systematic review. Int J Med Inform. 2023;180:105241. [DOI] [PubMed] [Google Scholar]
- 20.Binson VA, Thomas S, Subramoniam M, Arun J, Naveen S, Madhu S. A review of machine learning algorithms for biomedical applications. Ann Biomed Eng. 2024;52(5):1159–83. [DOI] [PubMed] [Google Scholar]
- 21.Elhaddad M, Hamam S. AI-driven clinical decision support systems: an ongoing pursuit of potential. Cureus. 2024;16(4):e57728. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.National Psoriasis Foundation. FY25-29 Strategic Plan. https://npf-website.cdn.prismic.io/npf-website/ZoKt1R5LeNNTwrl8_StrategicPlanFINAL.pdf. Accessed 14 Mar 2025.
- 23.Foster SA, Zhu B, Guo J, et al. Patient characteristics, health care resource utilization, and costs associated with treatment-regimen failure with biologics in the treatment of psoriasis. J Manag Care Spec Pharm. 2016;22(4):396–405. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Blauvelt A, Shi N, Burge R, et al. Comparison of real-world treatment patterns among biologic-experienced patients with psoriasis treated with ixekizumab or secukinumab over 18 months. Dermatol Ther Heidelb. 2021;11(6):2133–45. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Armstrong AW, Patel M, Li C, Garg V, Manava MR, Wu JJ. Real-world switching patterns and associated characteristics in patients with psoriasis treated with biologics in the United States. J Dermatolog Treat. 2023;34(1):2200870. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.