Characterizing patients with psoriasis on injectable biologics adalimumab, etanercept, and ustekinumab: A chart review study.

OBJECTIVE
This study examined plaque psoriasis (PsO) patient characteristics across injectable biologics.


METHODS
Data were collected from 400 US dermatologists randomly selecting five charts each for patients with PsO (patient n  =  2000): adalimumab (ADA; n  =  447), etanercept (ETA; 539), ustekinumab (UST) 45 mg (511) and UST 90 mg (503). Physicians had to have been in practice 2-30 years, managing 10+  patients (5 + with biologics for PsO). Generalized estimating equation models, weighted according to inverse probability of patient selection and accounting for patient correlation within physicians, examined patient measures as a function of treatment (UST 90 mg = reference).


RESULTS
Patients on UST 90 mg had higher odds of weighing  >100 kg (adjusted mean  =  34.4%) vs. ADA (10.9%), ETA (5.5%) or UST 45 mg (6.8%), greater body surface affected and higher odds of severe PsO prior to treatment and higher odds of prior biologics use. Mean prior biologics used was higher with UST 90 mg versus ADA or ETA. Number of comorbidities was higher with UST 90 mg versus ETA or UST 45 mg.


CONCLUSIONS
Among biologics-treated patients with PsO, UST 90 mg appears to be used in patients with greater weight, baseline severity and prior biologics experience than ADA, ETA or UST 45 mg. UST 90 mg is used in patients with more comorbidities than other treatments except ADA.


Introduction
Plaque psoriasis (PsO) is a chronic inflammatory, multisystem disorder with predominantly skin-and joint-related symptoms, affecting approximately 2% of the US population (1). About 20-30% of patients experience a moderate-to-severe condition that affects at least 5% of their body surface area and/or involves areas such as hands, feet, face or genitalia (1) and may require systemic treatments in addition to topical treatments to ameliorate inflammation (2).
Biologic agents have been found to be effective in treating moderate to severe PsO, while often requiring monitoring due to safety concerns or augmenting with combination or adjunctive therapies (3). At the time of the study, the most frequently used biologics in PsO were adalimumab (ADA) and etanercerpt (ETA), followed by ustekinumab (UST) and least frequently, infliximab, a biologic administered via infusion (4). This study focuses on the three injectable biologics: ADA, ETA and UST. Different from the other two injectable biologics, ustekinumab has a weightbased dosing, where the 90 mg is to be used in patients with a body weight greater than 100 kg and the 45 mg is to be used in patients weighing 100 kg or less (5).
However, little is known about how, at a population level, patients taking these biologics differ in terms of their demographic or health characteristics, including body weight. Therefore, research is needed to better understand whether patients with PsO using UST, particularly the 90 mg dose, differ from those using ADA and ETA in the real world. The objective of this study is to examine the characteristics of patient populations using each of the four biologics treatments (ADA, ETA, UST 45 mg and UST 90 mg), as well as to determine potential drivers of treatment choice.

Methods
Due to the large number of study measures, only measures of interest are reported in the main text. Complete methods are available in the online Appendix.

Sample and source
A total of 400 US dermatologists were asked to pull 2000 PsO patient charts (five per physician) split between those being treated with ADA, ETA, UST 45 mg and UST 90 mg. Participating dermatologists had to meet all of the following inclusion criteria to be eligible for enrollment into the study: (1) Dermatologists have the ability to read and write English.
(2) They should be a certified physician specializing in dermatology. (3) They should be board eligible or certified. (4) They have been practicing for 2-30 years, excluding residency and fellowship. (5) They must be managing at least 10 patients with psoriasis. (6) They must be managing at least five PsO patients with biologics.
(7) They must be treating at least 1 PsO patient with one of the treatments of interest to this analysis. (8) They have provided informed consent for participation in the study. Participating dermatologists were recruited from the All Global panel, an actively managed double opt-in online panel that closely matched the demographics of the American Medical Association (AMA) statistics with respect to region, age and gender (data not shown). All of the telephone recruitment calls are recorded, and at least 10% are randomly audited and monitored.
A two-stage sampling design was utilized to help ensure a convenience sample of dermatologists and a fairly representative selection of patients per treatment group among the sample of dermatologists, enabling later weighting of patients by their probability of selection in order to project accurately to the corresponding real-world patient population among the selected dermatologists. In the first stage, all dermatologists were contacted from a population of online physician panel members and invited to participate. In the second stage, each physician was asked to pull patient charts, randomly, within each of the requested treatment groups, for a total of five charts per physician.
No patient-identifying information was requested. The study was approved by Essex Institutional Review Board (Lebanon, NJ).

Measures
Information was collected from participating dermatologists who met the study inclusion criteria using a self-administered, internet-based questionnaire.

Sample description
The following physician-level variables were assessed to describe the convenience sample of physicians: gender, practice location, allocation of time spent on professional activities, ranking of biologics and frequency of prescription of biologics.

Patient-level measures
Patient sociodemographics and health characteristics included age, gender, ethnicity, height in feet, weight in pounds and number of comorbid conditions. These were extracted from patient charts and reported by participating dermatologists.
Healthcare access and treatment history variables included difficulty paying out-of-pocket (OOP) costs for biologics, treatments used prior to current treatment, current biologic used, prior biologic used and body coverage by plaque psoriasis prior to initiation of current biologic. These were also taken from patient charts and reported by participating dermatologists.
Additional potential predictors of treatment choice included the following variables, taken from patient charts by participating dermatologists and used in secondary analyses: employment status, characteristics of job (if employed), cigarette smoking, health insurance type, prescription medication coverage (if insured), time since diagnosis (in months), type of physician initiating treatment on current biologic, reasons for discontinuation (if patient switched biologics), patient weight during initiation decision, severity of plaque psoriasis prior to initiation, body locations covered prior to initiation, reasons for choosing current biologic over others, and mode of biologic administration.

Statistical analyses
Primary objective: understanding the characteristics of the patient populations of each treatment group Each treatment group was compared against the reference group (UST 90 mg). There was evidence of at least moderate intraclass correlations (ICCs) on some measures, and therefore, generalized estimating equation (GEE) models were used to examine individual patient characteristics (e.g. weight, age, gender) as a function of treatment groups (either ADA, ETA or UST 45 mg vs. UST 90 mg as the reference group). These models controlled simultaneously for any covariates; however, no covariates were entered for the ''adjusted bivariate'' analyses associated with the primary objective. These could be normal or binary logistic models, with the outputs providing Bs (betas) for normal distribution models or ORs (odds ratios) for binary logistic models, and estimated means/proportions and CIs (confidence intervals). GEE clusters were the groups of patients within physicians. The models were fit under the assumption of compound symmetry error structures.
Secondary objective: examining potential predictors of treatment choice Multivariable analysis was used to assess whether variables were independent predictors of treatment choice: three binary logistic GEE models predicting UST 90 mg versus either (1) UST 45 mg, (2) ETA or (3) ADA. As noted previously, with likely dependencies in the data, the multivariable model for the secondary objective was run with the same considerations as for the models used with the primary objective.
Covariates were selected according to several considerations (e.g. assessment of multicollinearity, preference for combined vs. low-prevalence and/or individual measures), with no more than 10-20 chosen for a given model. Although this was a crosssectional study, covariates were only selected if they were likely candidates as causal predictors (e.g. exogenous variables), not consequences (e.g. current disease severity), of the outcome measures (i.e. treatment group). Stepwise backward elimination was used with a simpler multinomial logistic GLM to help winnow the final remaining covariates list to a more manageable size including only significant contributors to the model fit.

Results
Due to the large number of study measures, only findings of interest were reported in the main text and tables. The complete results are available in the online Appendix.

Sample description
The final sample included 400 unique (sampled without replacement) dermatologists out of 11 821 invited from across two panels, with a 3.5% response rate (after accounting for 354 physicians who were screened out based on exclusion criteria). Study recruitment was terminated at 400 physicians; therefore, the overall potential response rate among invitees is unknown. Charts were pulled for 2000 patients in total: 447 on ADA (22.4%), 539 on ETA (27.0%), 511 on UST 45 mg (25.6%) and 503 on UST 90 mg (25.2%). Physicians were 66.3% male, 87.8% between 35-64 years old, 32.0% practicing in a major metropolitan area, and 51.5% in private group practice (results not shown). Physicians reported treating on average 91.6 (SD ¼ 112.7) psoriasis patients with biologics and spending on average 94.5% (SD ¼ 9.5) of their time in direct patient care. Physicians most frequently prescribed ADA as their typical first choice biologic therapy (46.0%), followed by ETA (39.3%) and UST (11.8%). Physicians most frequently prescribed UST as the third choice (51.5%). Physicians reported prescribing ADA to a mean 38.0% of their plaque PsO patients, with ETA prescribed to 31.6%, UST 45 mg to 16.0% and UST 90 mg to 12.0%. Patients were at higher odds of having used prior biologics with UST 90 mg (42.7%) than with ADA (18.3%), ETA (7.2%) and UST 45 mg (32.6%), all p50.01. Prior treatment was unknown for 41 patients (2.1%). Among 570 patients with known prior biologic experience (i.e. more than one type of prior biologic used), the mean was significantly higher with UST 90 mg (1.6) than with ADA (1.2) and ETA (1.2) (both p50.01) but not significantly higher than with UST 45 mg (1.4). A higher percentage of patients on UST 90 mg (16.9%) reported having difficulty paying out-of-pocket than patients on ETA (9.9%), p50.05 (Table 2).

Secondary objective: predictors of treatment choice
Adjusting for patient characteristics as described above, weight 4100 kg was a significant, independent predictor of UST 90 mg use versus UST 45 mg (odds ratio [OR] ¼ 3.8), ETA (OR ¼ 6.1) and ADA (OR ¼ 2.7), controlling for other variables, as was patient change in weight (OR ¼ 3.2, 7.0 and 6.3, respectively). Prior experience with biologics predicted UST 90 mg use versus ETA (OR ¼ 7.0 for 1 prior biologic and OR ¼ 19.6 for 2 + biologics) and ADA (OR ¼ 2.4 and OR ¼ 5.0, respectively). Prior moderate to severe (vs. clear to mild) PsO predicted UST 90 mg use versus ETA (OR ¼ 8.6). All p50.05 for above results (Table 3).
Convenient administration (OR ¼ 3.5), ease of administration (OR ¼ 5.1) better dosing schedule (OR ¼ 5.2), and faster improvement in symptoms (OR ¼ 3.3) were all significant reasons for choosing UST 90 mg versus ETA. Ease of administration (OR ¼ 2.4) was also a predictor of choosing UST 90 mg versus ADA. ''Would stop the progression of psoriatic arthritis'' was a predictor of choosing UST 90 mg versus UST 45 mg (OR ¼ 2.0) or ETA (OR ¼ 2.5). Ease of insurance approval, longer time on market and more experience with the current drug were significant predictors of choosing ETA (OR ¼ 4.3, 83.9 and 6.4, respectively) or ADA (OR ¼ 4.1, 23.7 and 4.9, respectively) over UST 90 mg. Not having difficulty paying out-of-pocket for prescription was a predictor of choosing ETA (OR ¼ 3.7) over UST 90 mg. Unknown coverage for prescription was a predictor of choosing UST 90 mg versus 45 mg (OR ¼ 2.9). Feet, toes and toenails, as locations affected prior to current treatment, was a predictor of UST 90 mg versus ADA use (OR ¼ 2.5). All p50.05 for the above results (Table 3).

Discussion
Among PsO patients treated with biologics, UST 90 mg was associated with greater comorbidities than ETA or UST 45 mg. Even after adjusting for covariates, UST 90 mg use was associated with higher weight, greater baseline severity, and more experience with prior biologics than ADA, ETA and/or UST 45 mg. The greater comorbidities observed among UST 90 mg users may reflect concomitant conditions experienced by overweight or obese patients who are the intended recipients of treatment according to weight-based guidelines. Although speculative, this point is supported by the absence of comorbidity as a significant predictor independent of patient weight in the predictive models. While previous research has assessed whether patient characteristics are associated with disease outcomes (6) and with treatment preferences (6)(7)(8), the current study is novel in describing and comparing the demographic and health characteristics of patients taking several biologic agents using a sampling framework that aims to represent patients in the real world.
For patients (as reported by participating dermatologists), convenient administration, ease of administration, better dosing schedule and faster improvement in symptoms were all significant reasons for choosing UST 90 mg over ETA. However, ease of insurance approval, longer time on market and more experience with the current drug were significant reasons for choosing ADA or ETA over UST 90 mg. These results corroborate prior research which has found that process-related factors, such as treatment location and delivery method, as well as disease outcomes, are important factors for patients in choosing treatment (8)(9). Notably, prior PsO treatment preference research has focused on nonbiologic treatments (7). The current study adds to the literature by describing patients among biologic agent treatment groups, which may be more relevant for clinical decision making. The adjustment for covariates in the predictive models also increases our understanding of potential drivers of treatment choice, as the significant predictors in these models reflect potential influences independent of the other measures (e.g. ease of administration remained a significant predictor of UST 90 mg vs. ADA and ETA, even after controlling for patients' experience with the drug and other variables that may be related).
Additionally, physicians' perceptions of the characteristics of each biologic were generally consistent with individual patient reasons (as reported by physicians) for choosing the particular biologic.

Strengths and limitations
One of the primary benefits of a physician chart review over, for example, patient-reported surveys or claims data, is that physicians are able to provide reliable information on patient characteristics (e.g. time since diagnosis, time since initiating treatment) for those on various treatments. Physician charts allowed us to access a fairly representative sample of patients (e.g. including patients who for various reasons would not choose to or be able to participate in an online survey), which is key for understanding patient characteristics associated with treatment use in the real world. Moreover, the internet-based format via which data were collected allowed for an expedient, cost-effective way to obtain extensive information from a large, geographically diverse sample. The current study revealed that physicians reported prescribing ADA and ETA to a higher proportion of their patients than UST (over twice as many). These results both highlighted the imbalance in prescriptions and therefore the need to oversample UST versus other biologics in the current study, plus they were used to generate the inverse proportional weights for analysis, thereby increasing the precision of within-group descriptive statistics and between-group comparisons.
Potential biases with patient charts include the following: (1) selection biases (e.g. physicians preferentially select charts for patients who are doing well on treatment); (2) physician selection (e.g. physicians may be selected who have preferences for one type of therapy over another); and (3) missing information on patient characteristics of interest (e.g. no data on patient education level or income, etc.). Participating dermatologists came from a panel of physicians largely representative of dermatologists in the United States, allowing for a variety of physician preferences and their corresponding patients. Patient chart selection biases were minimized via instructions to physicians (e.g. requests for pulling a randomly chosen qualifying chart for each treatment type). Certain characteristics were missing for many patients or on a physician-by-physician basis. This limitation is inherent in the design, but we made sure to capture as many related variables as possible to minimize this bias as well.
Other limitations of this study include that the data were reported by physicians, retrospectively and for a very small sample of each physician's patients (n ¼ 5). All effort was taken to avoid potential bias from this design, but it is impossible to exclude all sources of bias that could be introduced via this type of design. It is possible that the retrospective recall by physicians led to imprecise estimates of the reasons for prescribing different biologics by physicians. It is also difficult for us to control (beyond giving specific instructions) the specific charts that physicians pulled. However, in none of these cases, did we expect systematic bias on the part of the physician. In terms of missing information on patient characteristics of interest, a total of 414 patients had missing weight information. Additionally, there may be inaccurate recordings of patients' weight by physicians. The weight cutoff guidelines for UST 90 mg administration is technically 220 lbs.; however, an examination of the distribution of patients' weights revealed several modal responses, including a notable one at 200 lbs., suggesting that the likely cutoff for UST administration was in practice 200 lbs. However, it may also be the case that physicians were estimating patients' weights and used 200 lbs. as a convenient ''round number,'' when in fact those patients were 220lbs. or heavier.

Conclusions
This study reveals patient characteristics associated with dermatologists' prescribing behavior for biologics in PsO and potential drivers of treatment, which can help inform access and reimbursement decisions. In particular, patients with plaque PsO using UST 90 mg are on average different from those using ADA, ETA and UST 45 mg with respect to weight, prior severity and prior biologics experience. Future prospective studies are warranted to provide support for and replication of these findings.