Machine learning prediction and explanatory models of serious infections in patients with rheumatoid arthritis treated with tofacitinib.

BACKGROUND: Patients with rheumatoid arthritis (RA) have an increased risk of developing serious infections (SIs) vs. individuals without RA; efforts to predict SIs in this patient group are ongoing. We assessed the ability of different machine learning modeling approaches to predict SIs using baseline data from the tofacitinib RA clinical trials program. METHODS: This analysis included data from 19 clinical trials (phase 2, n = 10; phase 3, n = 6; phase 3b/4, n = 3). Patients with RA receiving tofacitinib 5 or 10 mg twice daily (BID) were included in the analysis; patients receiving tofacitinib 11 mg once daily were considered as tofacitinib 5 mg BID. All available patient-level baseline variables were extracted. Statistical and machine learning methods (logistic regression, support vector machines with linear kernel, random forest, extreme gradient boosting trees, and boosted trees) were implemented to assess the association of baseline variables with SI (logistic regression only), and to predict SI using selected baseline variables using 5-fold cross-validation. Missing values were handled individually per prediction model. RESULTS: A total of 8404 patients with RA treated with tofacitinib were eligible for inclusion (15,310 patient-years of total follow-up) of which 473 patients reported SIs. Amongst other baseline factors, age, previous infection, and corticosteroid use were significantly associated with SI. When applying prediction modeling for SI across data from all studies, the area under the receiver operating characteristic (AUROC) curve ranged from 0.656 to 0.739. AUROC values ranged from 0.599 to 0.730 in data from phase 3 and 3b/4 studies, and from 0.563 to 0.643 in data from ORAL Surveillance only. CONCLUSIONS: Baseline factors associated with SIs in the tofacitinib RA clinical trial program were similar to established SI risk factors associated with advanced treatments for RA. Furthermore, while model performance in predicting SI was similar to other published models, this did not meet the threshold for accurate prediction (AUROC > 0.85). Thus, predicting the occurrence of SIs at baseline remains challenging and may be complicated by the changing disease course of RA over time. Inclusion of other patient-associated and healthcare delivery-related factors and harmonization of the duration of studies included in the models may be required to improve prediction. TRIAL REGISTRATION: ClinicalTrials.gov: NCT00147498; NCT00413660; NCT00550446; NCT00603512; NCT00687193; NCT01164579; NCT00976599; NCT01059864; NCT01359150; NCT02147587; NCT00960440; NCT00847613; NCT00814307; NCT00856544; NCT00853385; NCT01039688; NCT02187055; NCT02831855; NCT02092467.

Impact of Previous Alopecia Areata Treatment on Efficacy Responses up to Week 48 Following Ritlecitinib Treatment: A Post Hoc Analysis.

INTRODUCTION: Patients with alopecia areata (AA) may have received several therapies for management of AA during their lives. In the ALLEGRO phase 2b/3 (NCT03732807) study, the oral JAK3/TEC family kinase inhibitor ritlecitinib demonstrated efficacy and an acceptable safety profile in patients aged ≥ 12 years with AA and ≥ 50% scalp hair loss. This post hoc analysis investigated associations between prior use of AA therapies and Severity of Alopecia Tool (SALT) responses in patients receiving ritlecitinib for AA. METHODS: Patients receiving ritlecitinib 30 mg or 50 mg once daily with or without an initial 4-week 200-mg daily loading dose were grouped by previous exposure to AA treatments, including topicals, intralesional corticosteroids (ILCS), topical immunotherapy, and systemic immunosuppressants or any prior AA treatment. Multivariable logistic regression analyses evaluated the association between response based on a SALT score of ≤ 20 and any prior treatment for AA at weeks 24 and 48. RESULTS: Of 522 patients, 360 (69.0%) had previous exposure to any AA treatment. At Week 24, SALT ≤ 20 response was positively associated with prior use of ILCS (odds ratio [OR], 2.12; 95% confidence interval [CI], 1.23-3.65; P < 0.05) and negatively associated with prior use of systemic immunosuppressants (OR 0.50; 95% CI 0.28-0.88; P < 0.05). Prior use of topicals or topical immunotherapy was not associated with SALT ≤ 20 response at Week 24. By Week 48, no association was identified between SALT ≤ 20 response and prior use of topicals, ILCS, topical immunosuppressants, or systemic immunosuppressants (all P > 0.05). Previous exposure to any AA therapy was not associated with SALT ≤ 20 response at weeks 24 or 48 (all P > 0.05). CONCLUSIONS: Prior AA treatment history had no effect on longer-term treatment response to ritlecitinib. TRIAL REGISTRATION NUMBER: NCT03732807.

A Two-Step, Trajectory-Focused, Analytics Approach to Attempt Prediction of Analgesic Response in Patients with Moderate-to-Severe Osteoarthritis.

INTRODUCTION: We sought to predict analgesic response to daily oral nonsteroidal anti-inflammatory drugs (NSAIDs) or subcutaneous tanezumab 2.5 mg (every 8 weeks) at week 16 in patients with moderate-to-severe osteoarthritis, based on initial treatment response over 8 weeks. METHODS: Data were derived from three randomized controlled trials of osteoarthritis. A two-step, trajectory-focused, analytics approach was used to predict patients as responders or non-responders at week 16. Step 1 identified patients using a data-element combination method (based on pain score at baseline, pain score at week 8, pain score monotonicity at week 8, pain score path length at week 8, and body site [knee or hip]). Patients who could not be identified in step 1 were predicted in step 2 using a k-nearest neighbor method based on pain score and pain response level at week 8. RESULTS: Our approach predicted response with high accuracy in NSAID-treated (83.2-90.2%, n = 931) and tanezumab-treated (84.6-91.0%, n = 1430) patients regardless of the efficacy measure used to assess pain, or the threshold used to define response (20%, 30%, or 50% improvement from baseline). Accuracy remained high using 50% or 20% response thresholds, with 50% and 20% yielding generally slightly better negative and positive predictive value, respectively, relative to 30%. Accuracy was slightly better in patients aged ≥ 65 years relative to younger patients across most efficacy measure/response threshold combinations. CONCLUSIONS: Analyzing initial 8-week analgesic responses using a two-step, trajectory-based approach can predict future response in patients with moderate-to-severe osteoarthritis treated with NSAIDs or 2.5 mg tanezumab. These findings demonstrate that prediction of treatment response based on a single dose of a novel therapeutic is possible and that predicting future outcomes based on initial response offers a way to potentially advance the approach to clinical management of patients with osteoarthritis. GOV IDENTIFIERS: NCT02528188, NCT02709486, NCT02697773.

Modeling of Treatment Outcomes with Tofacitinib Maintenance Therapy in Patients with Ulcerative Colitis: A Post Hoc Analysis of Data from the OCTAVE Clinical Program.

INTRODUCTION: Tofacitinib is an oral small molecule Janus kinase inhibitor for the treatment of ulcerative colitis (UC). This post hoc analysis assessed whether various statistical techniques could predict outcomes of tofacitinib maintenance therapy in patients with UC. METHODS: Data from patients who participated in a 52-week, phase III maintenance study (OCTAVE Sustain) and an open-label long-term extension study (OCTAVE Open) were included in this analysis. Patients received tofacitinib 5 or 10 mg twice daily (BID) or placebo (OCTAVE Sustain only). Logistic regression analyses were performed to generate models using clinical and laboratory variables to predict loss of responder status at week 8 of OCTAVE Sustain, steroid-free remission (defined as a partial Mayo score of 0-1 in the absence of corticosteroid use) at week 52 of OCTAVE Sustain, and delayed response at week 8 of OCTAVE Open. Furthermore, differences in loss of response/discontinuation patterns between treatment groups in OCTAVE Sustain were established. RESULTS: The generated prediction models demonstrated insufficient accuracy for determining loss of response at week 8, steroid-free remission at week 52 in OCTAVE Sustain, or delayed response in OCTAVE Open. Both tofacitinib doses demonstrated comparable response/remission patterns based on visualizations of disease activity over time. The rectal bleeding subscore was the primary determinant of disease worsening (indicated by an increased total Mayo score), and the endoscopy subscore was the primary determinant of disease improvement (indicated by a decreased total Mayo score). CONCLUSION: Visualizations of disease activity subscores revealed distinct patterns among patients with UC that had disease worsening and disease improvement. The statistical models assessed in this analysis could not accurately predict loss of responder status, steroid-free remission, or delayed response to tofacitinib. Possible reasons include the small sample size or missing data related to yet unknown key variables that were not collected during these trials.

Doctors use tofacitinib (Xeljanz^®) to treat people with moderate to severe ulcerative colitis. Patients who respond to (have improved symptoms following) treatment with tofacitinib 10 mg twice a day for 8 weeks, or up to 16 weeks if they do not respond initially (known as induction treatment), can receive tofacitinib treatment at the lowest effective dose to sustain their response (called maintenance treatment). Predicting how patients respond to tofacitinib maintenance treatment may help clinicians work out the lowest effective dose for each patient. In this study, data from the tofacitinib clinical trials were used to assess the ability to predict maintenance therapy response or failure in patients with ulcerative colitis. Differences between patients who received tofacitinib 5 or 10 mg twice a day and who either stopped responding to treatment or stopped taking treatment were looked at. The study could not accurately predict which patients would experience disease worsening, steroid-free remission (very mild or no symptoms, and not taking steroids), or take longer to respond following tofacitinib maintenance treatment. Patterns of patients who had stopped responding to treatment, or stopped taking treatment, were similar between patients who received tofacitinib 5 or 10 mg twice daily. When reviewed using doctor- and patient-reported scores that measure ulcerative colitis disease activity, different factors were important in patients with disease worsening compared with disease improvement. The results suggest that further research is needed to more accurately predict how patients with ulcerative colitis will respond to tofacitinib maintenance treatment.

Hair Loss Profiles and Ritlecitinib Efficacy in Patients with Alopecia Areata: Post Hoc Analysis of the ALLEGRO Phase 2b/3 Study.

INTRODUCTION: Ritlecitinib demonstrated efficacy in patients with alopecia areata (AA) in the ALLEGRO phase 2b/3 study (NCT03732807). However, hair loss presentation may vary based on location (e.g., scalp, eyebrow/eyelash, body). Here, we sought to identify distinct hair loss profiles at baseline and evaluate whether they affected the efficacy of ritlecitinib. METHODS: Patients with AA aged ≥ 12 years with ≥ 50% scalp hair loss were randomized to daily ritlecitinib 10 mg (assessed for dose ranging only), 30 or 50 mg (± 4-week, 200-mg loading dose), or placebo for 24 weeks. Latent class analysis (LCA) identified hair loss profiles based on four baseline measurements: clinician-reported extent of scalp (Severity of Alopecia Tool score), eyebrow hair loss, eyelash hair loss, and patient-reported body hair loss. Logistic regression evaluated ritlecitinib (50 and 30 mg) efficacy vs placebo using Patient Global Impression of Change (PGI-C) and Patient Satisfaction with Hair Growth (P-Sat; amount, quality, and overall satisfaction) responses at Week 24, adjusting for key covariates, including latent class membership. RESULTS: LCA identified five latent classes: (1) primarily non-alopecia totalis (AT; complete loss of scalp hair); (2) non-AT with moderate non-scalp involvement; (3) extensive scalp, eyebrow, and eyelash involvement; (4) AT with moderate non-scalp involvement; and (5) primarily alopecia universalis (complete scalp, face, and body hair loss). Adjusting for latent class membership, patients receiving ritlecitinib 30 or 50 mg were significantly more likely to achieve PGI-C response (30 mg: odds ratio, 8.62 [95% confidence interval, 4.42-18.08]; 50 mg: 12.29 [6.29-25.85]) and P-Sat quality of hair regrowth (30 mg: 6.71 [3.53-13.51]; 50 mg: 8.17 [4.30-16.46]) vs placebo at Week 24. Results were similar for P-Sat overall satisfaction and amount of hair regrowth. CONCLUSION: Distinct and clinically relevant hair loss profiles were identified in ALLEGRO-2b/3 participants. Ritlecitinib was efficacious compared with placebo, independent of hair loss profile at baseline. TRIAL REGISTRATION: ClinicalTrials.gov identifier, NCT03732807.

Characterizing 16-Week Responder Profiles Using Group-Based Trajectory Modeling in Over 4300 Clinical Trial Participants Receiving Pharmaceutical Treatment for Moderate to Severe Osteoarthritis.

INTRODUCTION: We sought to identify and characterize distinct responder profiles among osteoarthritis (OA) subjects treated with tanezumab, nonsteroidal anti-inflammatory drugs (NSAIDs), or placebo. METHODS: Subject-level data were derived from three randomized, double-blind, placebo- or NSAID-controlled trials of tanezumab in subjects with moderate-to-severe OA. Subjects received subcutaneous tanezumab (2.5 mg, n = 1527; 5 mg, n = 1279) every 8 weeks, oral NSAIDs (n = 994) daily, or placebo (n = 513). Group-based trajectory modeling (GBTM, an application of finite mixture statistical modeling that uses response trajectory to identify and summarize complex patterns in longitudinal data) was used to identify subgroups of subjects following similar patterns of response in each treatment arm, based on daily pain intensity scores from baseline through Week 16. We then examined whether subject-related variables were associated with any of the subgroups using multinomial logistic regression. RESULTS: A three-subgroup/four-inflection point trajectory model was selected based on clinical and statistical considerations. The subgroups were high responders (substantial pain improvement and a large majority of members achieved ≥ 30% improvement before Week 16), medium responders (gradual pain improvement and a majority of members achieved ≥ 30% improvement by Week 16), and non-responders (little to no pain improvement over 16 weeks). Across all treatments, fluctuation in pain intensity in the week prior to treatment was consistently associated with treatment response. Other variables were positively (age, body mass index, days of rescue medication use) or negatively (severity of disease based on Kellgren-Lawrence grading) associated with response but effects were small and/or varied across treatments. CONCLUSIONS: Across all treatments, GBTM identified three subgroups of subjects that were characterized by extent of treatment response (high, medium, and non-responders). Similar analyses (e.g., grouping of subjects based on response trajectory and identification of subgroup-related variables) in other studies of OA could inform clinical trial design and/or treatment approaches. (NCT02697773; NCT02709486; NCT02528188).

Prediction of early clinical response in patients receiving tofacitinib in the OCTAVE Induction 1 and 2 studies.

INTRODUCTION: Tofacitinib is an oral, small molecule Janus kinase inhibitor for the treatment of ulcerative colitis (UC). Outcome prediction based on early treatment response, along with clinical and laboratory variables, would be very useful for clinical practice. The aim of this study was to determine early variables predictive of responder status in patients with UC treated with tofacitinib. METHODS: Data were collected from patients treated with tofacitinib 10 mg twice daily in the OCTAVE Induction 1 and 2 studies (NCT01465763 and NCT01458951). Logistic regression and random forest analyses were performed to determine the power of clinical and/or laboratory variables to predict 2- and 3-point partial Mayo score responder status of patients at Weeks 4 or 8 after baseline. RESULTS: From a complete list of variables measured in OCTAVE Induction 1 and 2, analyses identified partial Mayo score, partial Mayo subscore (stool frequency, rectal bleeding, and Physician Global Assessment), cholesterol level, and C-reactive protein level as sufficient variables to predict responder status. Using these variables at baseline and Week 2 predicted responder status at Week 4 with 84-87% accuracy and Week 8 with 74-79% accuracy. Variables at baseline, Weeks 2 and 4 could predict responder status at Week 8 with 85-87% accuracy. CONCLUSION: Using a limited set of time-dependent variables, statistical and machine learning models enabled early and clinically meaningful predictions of tofacitinib treatment outcomes in patients with moderately to severely active UC.

Correction: Using time series analysis approaches for improved prediction of pain outcomes in subgroups of patients with painful diabetic peripheral neuropathy.

[This corrects the article DOI: 10.1371/journal.pone.0207120.].

Integrating Machine Learning With Microsimulation to Classify Hypothetical, Novel Patients for Predicting Pregabalin Treatment Response Based on Observational and Randomized Data in Patients With Painful Diabetic Peripheral Neuropathy.

PURPOSE: Variability in patient treatment responses can be a barrier to effective care. Utilization of available patient databases may improve the prediction of treatment responses. We evaluated machine learning methods to predict novel, individual patient responses to pregabalin for painful diabetic peripheral neuropathy, utilizing an agent-based modeling and simulation platform that integrates real-world observational study (OS) data and randomized clinical trial (RCT) data. PATIENTS AND METHODS: The best supervised machine learning methods were selected (through literature review) and combined in a novel way for aligning patients with relevant subgroups that best enable prediction of pregabalin responses. Data were derived from a German OS of pregabalin (N=2642) and nine international RCTs (N=1320). Coarsened exact matching of OS and RCT patients was used and a hierarchical cluster analysis was implemented. We tested which machine learning methods would best align candidate patients with specific clusters that predict their pain scores over time. Cluster alignments would trigger assignments of cluster-specific time-series regressions with lagged variables as inputs in order to simulate "virtual" patients and generate 1000 trajectory variations for given novel patients. RESULTS: Instance-based machine learning methods (k-nearest neighbor, supervised fuzzy c-means) were selected for quantitative analyses. Each method alone correctly classified 56.7% and 39.1% of patients, respectively. An "ensemble method" (combining both methods) correctly classified 98.4% and 95.9% of patients in the training and testing datasets, respectively. CONCLUSION: An ensemble combination of two instance-based machine learning techniques best accommodated different data types (dichotomous, categorical, continuous) and performed better than either technique alone in assigning novel patients to subgroups for predicting treatment outcomes using microsimulation. Assignment of novel patients to a cluster of similar patients has the potential to improve prediction of patient outcomes for chronic conditions in which initial treatment response can be incorporated using microsimulation. CLINICAL TRIAL REGISTRIES: www.clinicaltrials.gov: NCT00156078, NCT00159679, NCT00143156, NCT00553475.

Predicting Responses to Pregabalin for Painful Diabetic Peripheral Neuropathy Based on Trajectory-Focused Patient Profiles Derived from the First 4 Weeks of Treatment.

INTRODUCTION: Prediction of final clinical outcomes based on early weeks of treatment can enable more effective patient care for chronic pain. Our goal was to predict, with at least 90% accuracy, 12- to 13-week outcomes for pregabalin-treated painful diabetic peripheral neuropathy (pDPN) patients based on 4 weeks of pain and pain-related sleep interference data. METHODS: We utilized active treatment data from six placebo-controlled randomized controlled trials (n = 939) designed to evaluate efficacy of pregabalin for reducing pain in patients with pDPN. We implemented a three-step, trajectory-focused analytics approach based upon patient responses collected during the first 4 weeks using monotonicity, path length, frequency domain (FD), and k-nearest neighbor (kNN) methods. The first two steps were based on combinations of baseline pain, pain at 4 weeks, weekly monotonicity and path length during the first 4 weeks, and assignment of patients to one of four responder groups (based on presence/absence of 50% or 30% reduction from baseline pain at 4 and at 12/13 weeks). The third step included agreement between prediction of logistic regression of daily FD amplitudes and assignment made from kNN analyses. RESULTS: Step 1 correctly assigned 520/939 patients from the six studies to a responder group using a 3-metric combination approach based on unique assignment to a 50% responder group. Step 2 (applied to the remaining 419 patients) predicted an additional 121 patients, using a blend of 50% and 30% responder thresholds. Step 3 (using a combination of FD and kNN analyses) predicted 204 of the remaining 298 patients using the 50% responder threshold. Our approach correctly predicted 90.0% of all patients. CONCLUSION: By correctly predicting 12- to 13-week responder outcomes with 90% accuracy based on responses from the first month of treatment, we demonstrated the value of trajectory measures in predicting pDPN patient response to pregabalin. TRIAL REGISTRATION: www.clinicaltrials.gov identifiers, NCT00156078/NCT00159679/NCT00143156/NCT00553475. FUNDING: Pfizer. Plain language summary available for this article.

Dose Titration of Pregabalin in Patients with Painful Diabetic Peripheral Neuropathy: Simulation Based on Observational Study Patients Enriched with Data from Randomized Studies.

INTRODUCTION: Achieving a therapeutic response to pregabalin in patients with painful diabetic peripheral neuropathy (pDPN) requires adequate upward dose titration. Our goal was to identify relationships between titration and response to pregabalin in patients with pDPN. METHODS: Data were integrated from nine randomized, placebo-controlled clinical trials as well as one 6-week open-label observational study conducted by 5808 physicians (2642 patients with pDPN) in standard outpatient settings in Germany. These studies evaluated pregabalin for treatment of pDPN. Using these data, we examined "what if" scenarios using a microsimulation platform that integrates data from randomized and observational sources as well as autoregressive-moving-average with exogenous inputs models that predict pain outcomes, taking into account weekly changes in pain, sleep interference, dose, and other patient characteristics that were unchanging. RESULTS: Final pain levels were significantly different depending on dose changes (P < 0.0001), with greater proportions improving with upward titration regardless of baseline pain severity. Altogether, 78.5% of patients with pDPN had 0-1 dose change, and 15.2% had ≥ 2 dose changes. Simulation demonstrated that the 4.8% of inadequately titrated patients who did not improve/very much improve their pain levels would have benefited from ≥ 2 dose changes. Patient satisfaction with tolerability (range 90.3-96.2%) was similar, regardless of baseline pain severity, number of titrations, or extent of improvement, suggesting that tolerability did not influence treatment response patterns. CONCLUSION: Upward dose titration reduced pain in patients with pDPN who actually received it. Simulation also predicted pain reduction in an inadequately titrated nonresponder subgroup of patients had they actually received adequate titration. The decision not to uptitrate must have been driven by factors other than tolerability. FUNDING: Pfizer, Inc.

Using time series analysis approaches for improved prediction of pain outcomes in subgroups of patients with painful diabetic peripheral neuropathy.

Prior work applied hierarchical clustering, coarsened exact matching (CEM), time series regressions with lagged variables as inputs, and microsimulation to data from three randomized clinical trials (RCTs) and a large German observational study (OS) to predict pregabalin pain reduction outcomes for patients with painful diabetic peripheral neuropathy. Here, data were added from six RCTs to reduce covariate bias of the same OS and improve accuracy and/or increase the variety of patients for pain response prediction. Using hierarchical cluster analysis and CEM, a matched dataset was created from the OS (N = 2642) and nine total RCTs (N = 1320). Using a maximum likelihood method, we estimated weekly pain scores for pregabalin-treated patients for each cluster (matched dataset); the models were validated with RCT data that did not match with OS data. We predicted novel 'virtual' patient pain scores over time using simulations including instance-based machine learning techniques to assign novel patients to a cluster, then applying cluster-specific regressions to predict pain response trajectories. Six clusters were identified according to baseline variables (gender, age, insulin use, body mass index, depression history, pregabalin monotherapy, prior gabapentin, pain score, and pain-related sleep interference score). CEM yielded 1766 patients (matched dataset) having lower covariate imbalances. Regression models for pain performed well (adjusted R-squared 0.90-0.93; root mean square errors 0.41-0.48). Simulations showed positive predictive values for achieving >50% and >30% change-from-baseline pain score improvements (range 68.6-83.8% and 86.5-93.9%, respectively). Using more RCTs (nine vs. the earlier three) enabled matching of 46.7% more patients in the OS dataset, with substantially reduced global imbalance vs. not matching. This larger RCT pool covered 66.8% of possible patient characteristic combinations (vs. 25.0% with three original RCTs) and made prediction possible for a broader spectrum of patients. Trial Registration: www.clinicaltrials.gov (as applicable): NCT00156078, NCT00159679, NCT00143156, NCT00553475.