New Insights Into the Therapeutic Management of Varicella Zoster Virus Meningitis: A Series of 123 Polymerase Chain Reaction-Confirmed Cases.

Background: Varicella zoster virus (VZV) can reactivate and cause meningitis, but few studies have distinguished it from meningoencephalitis regarding treatment recommendations.The objective of this study was to assess the outcomes of a large series of patients with VZV meningitis according to their therapeutic management. Methods: We conducted a bicentric retrospective cohort study, in Paris, France, including all adult patients with a cerebrospinal fluid sample positive for VZV by polymerase chain reaction between April 2014 and June 2022. We distinguished meningitis from encephalitis according to the International Encephalitis Consortium criteria. Unfavorable outcome was defined as mortality or functional sequelae defined by a loss of 2 points on the modified Rankin Scale. Results: We included 123 patients with meningitis. Among them, 14% received no antivirals, while 20% were treated with oral valacyclovir alone, 41% with a short course of intravenous (IV) acyclovir before switch to valacyclovir, and 25% with a long course of IV acyclovir. Outcomes were favorable regardless of antiviral regimen. In multivariate analysis, only age, underlying immunosuppression, and cranial radiculitis appear to be predictive factors for longer IV therapy, based on the Akaike information criterion. Conclusions: In this study, patients with VZV meningitis had a good outcome, with no evidence of any impact of the treatment strategy. However, further studies are needed to support the possibility of milder treatment in immunocompetent patients, avoiding cost and side effects of IV acyclovir.

Two-stage randomized clinical trials with a right-censored endpoint: Comparison of frequentist and Bayesian adaptive designs.

Adaptive randomized clinical trials are of major interest when dealing with a time-to-event outcome in a prolonged observation window. No consensus exists either to define stopping boundaries or to combine p $$ p $$ values or test statistics in the terminal analysis in the case of a frequentist design and sample size adaptation. In a one-sided setting, we compared three frequentist approaches using stopping boundaries relying on α $$ \alpha $$ -spending functions and a Bayesian monitoring setting with boundaries based on the posterior distribution of the log-hazard ratio. All designs comprised a single interim analysis with an efficacy stopping rule and the possibility of sample size adaptation at this interim step. Three frequentist approaches were defined based on the terminal analysis: combination of stagewise statistics (Wassmer) or of p $$ p $$ values (Desseaux), or on patientwise splitting (Jörgens), and we compared the results with those of the Bayesian monitoring approach (Freedman). These different approaches were evaluated in a simulation study and then illustrated on a real dataset from a randomized clinical trial conducted in elderly patients with chronic lymphocytic leukemia. All approaches controlled for the type I error rate, except for the Bayesian monitoring approach, and yielded satisfactory power. It appears that the frequentist approaches are the best in underpowered trials. The power of all the approaches was affected by the violation of the proportional hazards (PH) assumption. For adaptive designs with a survival endpoint and a one-sided alternative hypothesis, the Wassmer and Jörgens approaches after sample size adaptation should be preferred, unless violation of PH is suspected.

Risk factors and outcome associated with fungal infections in patients with severe burn injury: 10-year retrospective IFI-BURN study.

BACKGROUND: In burn patients, skin barrier disruption and immune dysfunctions increase susceptibility to invasive fungal diseases (IFDs) like invasive candidiasis (IC) and invasive mold infections (IMI). We provide an in-depth analysis of IFD-related factors and outcomes in a 10-year cohort of severe burn patients. METHOD: Retrospective cohort study including adult patients admitted to the Burn Intensive Care Unit (BICU) between April 2014 and May 2023 with Total Burn Surface Area (TBSA) ≥15%. Patients were classified as proven IFD according to EORTC/MSGERC criteria applicable for IC. Putative IMIs were defined with: ≥2 positive cultures from a skin biopsy/bronchoalveolar lavage OR ≥2 positive blood specific-qPCRs OR a combination of both. RESULTS: Among 1381 patients admitted, 276 consecutive patients with TBSA ≥15% were included. Eighty-seven (31.5%; IC n=30; IMI n=43; both n=14) patients fulfilled the criteria for probable/putative IFD. At Day 30 after the burn injury, the estimated cumulative incidence pr/pu IFD was 26.4% (95%CI 21.4-31.8%). Factors independently associated with IFDs were TBSA, severity scores and indoor burn injury (i.e., from confined space fire). Overall mortality was 15.3% and 36.8% in the no IFD, pr/pu IFD groups respectively (p<0.0001). IFD was independently associated with a risk of death (HR: 1.94 for pr/pu IFD; 95%CI, 1.12-3.36; p=0.019). DISCUSSION: This study describes 21st-century characteristics of IFDs in sever burn patients confirming known risk factors with thresholds and identifying the indoor injury as an independent factor associated to IFDs. This suggests a link to contamination caused by fire damage, which is highly susceptible to aerosolizing spores.

Low cabotegravir trough concentrations without oral lead-in in patients with HIV-1 switching to long-acting cabotegravir and rilpivirine.

In a cohort of 72 consecutive virologically-suppressed patients with HIV-1 switching to long-acting cabotegravir and rilpivirine, we observed low cabotegravir trough concentrations 1 and 3 months after the first injection, with a significant association with no oral lead-in at 1 month [odds ratio (OR) = 6.3 [95% confidence interval (CI) 1.7-29.5], P = 0.01] and three months (OR = 5.6 [95% CI 1.3-29.7], P = 0.03), and with high BMI at 1 month (OR = 1.3 [95% CI 1.1-1.6], P = 0.007).

Composite event-free-survival as an endpoint in oncology drug evaluation: Review and guidance perspectives from the Haute Autorité de Santé (HAS).

BACKGROUND: The use of right-censored composite endpoints, such as progression-free survival, has been questioned in haemato-oncology trials due to potential bias in estimated treatment effect. This may impact the accuracy of health technology evaluations. We hypothesized that there is heterogeneity and potential sources of bias in the reporting of composite endpoints to health technology assessment (HTA) bodies. METHODS: We reviewed the submissions for reimbursement of oncology drugs in 2021 and 2022 that used a composite endpoint in the pivotal trial, after appraisal by the French HTA body. The retrieved information included the clinical study report, protocol, and statistical analysis plan submitted by the industry. All events of the composite endpoint and all causes of censored observations were measured. The design characteristics and treatment effect estimates were recorded. FINDINGS: Seventy-six submissions were selected, including seven without a right-censored endpoint and four evaluating associations, resulting in 65 analysed records: 17 for haematological and 48 for solid tumours. Out these 65 submissions, 47 (72·3%) used a randomized controlled design, and 18 (27·7%) a non-comparative design. The most frequently used composite endpoint was progression-free survival, used in 54 (83·1%) of the submissions. Censoring was possibly informative in 51 (92·7%) cases, mostly due to the onset of new treatment (44/51, 86·3%) and/or discontinuation of follow-up (33/51, 64·7%). In contrast, 38 (58·5%) trials reported a quantification of censored observations, with only 12/51 (23·5%) quantifying the informative ones. The estimated treatment effect on the composite outcome increased with the amount of censoring, suggesting a higher benefit of the drug, but remained below that on survival with poor evidence of surrogacy (R-squared=0·23). INTERPRETATION: Clinical study reports should be improved in terms of reporting censoring, while stakeholders should be aware of this potential source of bias. At a minimum, sensitivity analysis that ignores intercurrent events should be requested.

Transplantation for myelofibrosis patients in the ruxolitinib era: a registry study from the Société Francophone de Greffe de Moelle et de Thérapie Cellulaire.

In this SFGM-TC registry study, we report the results after stem cell transplantation (HSCT) in 305 myelofibrosis patients, in order to determine potential risk factors associated with outcomes, especially regarding previous treatment with ruxolitinib. A total of 102 patients were transplanted from an HLA-matched-sibling donor (MSD), and 143 patients received ruxolitinib. In contrast with previous studies, our results showed significantly worse outcomes for ruxolitinib patients regarding overall survival (OS) and non-relapse mortality (NRM), especially in the context of unrelated donors (URD). When exploring reasons for potential confounders regarding the ruxolitinib effect, an interaction between the type of donor and the use of ATG was found, therefore subsequent analyses were performed separately for each type of donor. Multivariable analyses did not confirm a significant negative impact of ruxolitinib in transplantation outcomes. In the setting of URD, only age and Fludarabine-Melphalan (FM) conditioning were associated with increased NRM. For MSD, only Karnoksfy <70% was associated with reduced OS. However, a propensity score analysis showed that ruxolitinib had a negative impact on OS but only in non-responding patients, consistent with previous data. To conclude, with all the precautions due to confounders and bias, ruxolitinib itself does not appear to increase mortality after HSCT.

Single-arm phase 3 designs: An oxymoron?

BACKGROUND: Since the 1950s, randomized clinical trials (RCTs) have served as the gold standard for confirming the benefits of a new drug. Accordingly, phase 3 trials, the last steps in the evaluation process for a new drug, have been recommended to all be RCTs. Nevertheless, single-arm phase 3 trials still appear to be in use. METHODS: We performed a PubMed search to identify the use of a single-arm design in phase 3 trials, excluding only non-English articles. Three categories were distinguished: past use of an RCT, of any other trial design, or no previous trial; and according to diagnosis (oncology, infection, others). RESULTS: A total of 176 single-arm phase 3 trials (19 oncology, 43 infections and 114 others) were identified by the search, with exponential growth since 1994, in parallel with that of RCTs. Among them, 64 (36%) were preceded by an RCT, 58 (33%) by a non-randomized trial, and 54 (31%) had no previous trial, with no main influence of the diagnosis setting. Justification of the design was reported in 30 (18%) of those trials, with ethical concerns comprising one-third of those justifications. This was similar in the 14 single-arm phase 2-3 trials, with about one-third in each group, and 17% justification of a non-comparative design. CONCLUSION: The use of a single-arm phase 3 trial is heterogeneous, ranging from first trials up to confirmatory trials after a previously conducted RCT. Justification for these single-arm designs as confirmatory evidence should be more clearly reported, along with potential sources of bias.

Nilotinib with or without cytarabine for Philadelphia-positive acute lymphoblastic leukemia.

ABSTRACT: We previously demonstrated that a reduced-intensity chemotherapy schedule can safely replace hyper-CVAD (cyclophosphamide-vincristine-doxorubicin [Adriamycin]-dexamethasone) cycle 1 when combined with imatinib in adults with Philadelphia-positive acute lymphoblastic leukemia. In the present randomized GRAAPH-2014 trial, we used nilotinib and addressed the omission of cytarabine (Ara-C) in consolidation. The primary objective was the major molecular response (MMR) rate measured by BCR::ABL1 quantification after cycle 4 (end of consolidation). All patients were eligible for allogeneic stem cell transplant (SCT), whereas those in MMR could receive autologous SCT, followed by 2-year imatinib maintenance in both cases. After the enrollment of 156 of 265 planed patients, the data and safety monitoring board decided to hold the randomization because of an excess of relapse in the investigational arm. Among the 155 evaluable patients, 76 received Ara-C during consolidation (arm A) and 79 did not (arm B). Overall, 133 patients (85%) underwent SCT, 93 allogeneic and 40 autologous. The noninferiority end point regarding MMR was reached with 71.1% (arm A) and 77.2% (arm B) of patients reaching MMR. However, the 4-year cumulative incidence of relapse was higher in arm B compared with arm A (31.3% [95% confidence interval {CI}, 21.1%-41.9%] vs 13.2% [95% CI, 6.7%-21.9%]; P = .017), which translated to a lower relapse-free survival. With a median follow-up of 3.8 years, 4-year overall survival was 79.0% (95% CI, 70.6%-89.3%) in arm A vs 73.4% (95% CI, 63.9%-84.4%) in arm B (P = .35). Despite a noninferior rate of MMR, more relapses were observed when ARA-C was omitted without impact on survival. ClinicalTrials.gov ID, NCT02611492.

Vaginal misoprostol versus vaginal dinoprostone for cervical ripening and induction of labour: An individual participant data meta-analysis of randomised controlled trials.

BACKGROUND: Induction of labour (IOL) is common practice and different methods carry different effectiveness and safety profiles. OBJECTIVES: To compare the effectiveness, and maternal and perinatal safety outcomes of IOL with vaginal misoprostol versus vaginal dinoprostone using individual participant data from randomised clinical trials. SEARCH STRATEGY: The following databases were searched from inception to March 2023: CINAHL Plus, ClinicalTrials.gov, Cochrane Pregnancy and Childbirth Group Trial Register, Ovid Embase, Ovid Emcare, Ovid MEDLINE, Scopus and the World Health Organization (WHO) International Clinical Trials Registry Platform (ICTRP). SELECTION CRITERIA: Randomised controlled trials (RCTs), with viable singleton gestation, no language restrictions, and all published and unpublished data. DATA COLLECTION AND ANALYSIS: An individual participant data meta-analysis was carried out. MAIN RESULTS: Ten of 52 eligible trials provided individual participant data, of which two were excluded after checking data integrity. The remaining eight trials compared low-dose vaginal misoprostol versus dinoprostone, including 4180 women undergoing IOL, which represents 32.8% of all participants in the published RCTs. Of these, 2077 were assigned to low-dose vaginal misoprostol and 2103 were assigned to vaginal dinoprostone. Compared with vaginal dinoprostone, low-dose vaginal misoprostol had a comparable rate of vaginal birth. Composite adverse perinatal outcomes did not differ between the groups. Compared with vaginal dinoprostone, composite adverse maternal outcomes were significantly lower with low-dose vaginal misoprostol (aOR 0.80, 95% CI 0.65-0.98, P = 0.03, I2 = 0%). CONCLUSIONS: Low-dose vaginal misoprostol and vaginal dinoprostone for IOL are comparable in terms of effectiveness and perinatal safety. However, low-dose vaginal misoprostol is likely to lead to a lower rate of composite adverse maternal outcomes than vaginal dinoprostone.

[Chronology of the ISO 9001:2015 certification and satisfaction survey in the multidisciplinary Clinical Investigation Center (CIC) of the historic Saint-Louis Hospital during the COVID-19 pandemic]. / Chronologie de la certification ISO 9001:2015 et enquête de satisfaction au sein du centre d'investigations cliniques plurithématique (CIC-P) de l'historique hôpital Saint-Louis en période de pandémie du COVID-19.

The main objectives of multidisciplinary clinical investigation center (CIC-P) are to facilitate the availability of new drugs for patients, to enhance the visibility and attractiveness of French clinical research, to improve the quality of early phase trials, and to enhance the value of academic research by evaluating molecules in rare diseases. Since 2017, the CIC-P has been committed to a quality approach process, launching in 2018 its first satisfaction survey on patient care and clinical trial management of all its employees. A second satisfaction survey targeted by profession type was to be launched in 2020, in view of the requirements of the ISO 9001:2015 standard, but the process was interrupted following the coronavirus diseases 2019 (COVID-19) pandemic. The successful reorganization of the CIC-P activity during the first containment of the COVID-19 pandemic was driven by the implementation of a quality management system that promotes continuous improvement through the organization and involvement of all the staff. This voluntary and participative approach motivated the CIC-P to apply for the organizational sesame. The ISO 9001:2015 certification of CIC-P aims at increasing its performance, to satisfying its customers and to fully integrate its activities in a continuous improvement process, according to the requirements of this international standard, through the deployment of quality tools such as The Deming wheel (PDCA), an indispensable tool for transformation and reorganization; the analysis of the environment by the strengths, weakness, opportunities, threats (SWOT) analysis tool; the analysis and management of risks by the FMEA method, and all with performance indicators (SMART) and precise objectives at each stage of a project/process. The implementation of satisfaction questionnaires remains the essential tool for evaluating the expectations and needs of interested parties, but also for improving the quality of CIC-P activities and services. All these tools put in place have allowed us to continuously improve the means of production and to constantly improve our organization.

Factors Associated With Treatment Pathways in Early Axial Spondyloarthritis: A Multistate Analysis of the 10-Year Follow-Up of the DESIR Cohort.

OBJECTIVE: Current recommendations for the management of patients with axial spondyloarthritis (axSpA) emphasize the need of an individualized strategy in therapeutic decision-making. The study objectives were to describe therapeutic strategies observed in axSpA, and to assess the factors associated with treatment intensification over time. METHODS: We included patients with axSpA from the French prospective cohort DESIR (Devenir des Spondylarthropathies Indifférenciées Récentes), with a scheduled 10-year follow-up. A multistate model with 4 ordered treatment states (no treatment, nonsteroidal antiinflammatory drugs [NSAIDs], conventional synthetic disease-modifying antirheumatic drugs [csDMARDs], and tumor necrosis factor inhibitors [TNFi]) was defined, with 6 possible transitions. Restricted mean sojourn times in each state were estimated. Then, predictors of those transitions were assessed by multivariable Cox models. RESULTS: A total of 686/708 (96.9%) patients who had > 1 visit were analyzed. At cohort entry, 199 (29%) were untreated, 427 (62.2%) were receiving NSAIDs, 60 (8.7%) csDMARDs, and none were receiving TNFi. Over the follow-up period, patients mostly (46.4% of the time) received NSAIDs, followed by TNFi (24.4% of the time). The presence of sacroiliitis on radiographs, inflammatory bowel disease, and articular index were jointly associated with the transition to NSAIDs. Longer duration in the previous state often decreased the hazard of the transition to csDMARDs or TNFi. Worse disease activity outcomes increased the hazard of most transitions. CONCLUSION: To our knowledge, this was the first study using a multistate model to easily represent different treatment states, detailing the transitions across them and their associated factors. Different time profiles for the management of patients with axSpA were identified, including in those abstaining from treatment up to a significant proportion of patients treated with csDMARDs.

A predictive classifier of poor prognosis in transplanted patients with juvenile myelomonocytic leukemia: a study on behalf of the Société Francophone de Greffe de Moelle et de Thérapie Cellulaire.

Juvenile myelomonocytic leukemia (JMML) is an aggressive pediatric myeloproliferative neoplasm requiring hematopoietic stem cell transplantation (HSCT) in most cases. We retrospectively analyzed 119 JMML patients who underwent first allogeneic HSCT between 2002 and 2021. The majority (97%) carried a RAS-pathway mutation, and 62% exhibited karyotypic alterations or additional mutations in SETBP1, ASXL1, JAK3 and/or the RAS pathway. Relapse was the primary cause of death, with a 5-year cumulative incidence of 24.6% (95%CI: 17.1-32.9). Toxic deaths occurred in 12 patients, resulting in treatmentrelated mortality (TRM) of 9.0% (95%CI: 4.6-15.3). The 5-year overall (OS) and event-free survival were 73.6% (95%CI: 65.7-82.4) and 66.4% (95%CI: 58.2-75.8), respectively. Four independent adverse prognostic factors for OS were identified: age at diagnosis >2 years, time from diagnosis to HSCT >6 months, monocyte count at diagnosis >7.2x109/L, and the presence of additional genetic alterations. Based on these factors, we proposed a predictive classifier. Patients with three or more predictors (21% of the cohort) had a 5-year OS of 34.2%, whereas those with none (7%) had a 5-year OS of 100%. Our study demonstrates improved transplant outcomes compared to prior published data, which can be attributed to the synergistic impacts of a low TRM and a reduced yet still substantial relapse incidence. By integrating genetic information with clinical and hematological features, we have devised a predictive classifier. This classifier effectively identifies a subgroup of patients who are at a heightened risk of unfavorable post-transplant outcomes who would benefit novel therapeutic agents and post-transplant strategies.

Early complete response as a validated surrogate marker in extranodal marginal zone lymphoma systemic therapy.

ABSTRACT: Extranodal marginal zone lymphoma (EMZL) has a very indolent course, and the validation of surrogate markers could accelerate novel therapies. Although prognostic markers do exist, no surrogate markers have been validated in EMZL. We hypothesized that time to complete response within 24 months (TTCR24) and complete response (CR) at 24 months (CR24) could be valid surrogate markers of progression-free survival (PFS). The International Extranodal Lymphoma Study Group 19 phase 3 trial showed the advantage of double therapy (rituximab + chlorambucil) over single therapy (rituximab or chlorambucil) on PFS. We used 2 recently published single-trial approaches to assess whether TTCR24 and CR24 were good surrogate markers of 8-year PFS (8y-PFS). Among the 401 patients, 264 (66%) reached a CR in the first 24 months, of which 222 (84%) remained in CR at month 24. The cumulative incidence of CR over time was significantly higher in patients under double therapy (hazard ratio, 1.75; P < .001). The double therapy arm was associated with a higher CR24 rate, a shorter TTCR24, and a longer 8y-PFS. The estimated proportion of treatment effect on 8y-PFS explained by TTCR24 was 95% (95% confidence interval [CI], 0.27-1.87). CR24 was also a strong surrogate marker because it mediated 90% (95% CI, 0.51-2.22) of the treatment effect on PFS and its natural indirect effect was significant throughout the follow-up. We found that TTCR24 predicted 95% and that CR24 mediated 90% of the treatment effect on long-term PFS. Therefore, TTCR24 and CR24 could be used in clinical trials as informative and valid early indicators of treatment effect on PFS. This trial was registered at www.clinicaltrials.gov as #NCT00210353.

Comparison of models for stroke-free survival prediction in patients with CADASIL.

Cerebral autosomal dominant arteriopathy with subcortical infarcts and leukoencephalopathy, which is caused by mutations of the NOTCH3 gene, has a large heterogeneous progression, presenting with declines of various clinical scores and occurrences of various clinical event. To help assess disease progression, this work focused on predicting the composite endpoint of stroke-free survival time by comparing the performance of Cox proportional hazards regression to that of machine learning models using one of four feature selection approaches applied to demographic, clinical and magnetic resonance imaging observational data collected from a study cohort of 482 patients. The quality of the modeling process and the predictive performance were evaluated in a nested cross-validation procedure using the time-dependent Brier Score and AUC at 5 years from baseline, the former measuring the overall performance including calibration and the latter highlighting the discrimination ability, with both metrics taking into account the presence of right-censoring. The best model for each metric was the componentwise gradient boosting model with a mean Brier score of 0.165 and the random survival forest model with a mean AUC of 0.773, both combined with the LASSO feature selection method.

Respiratory torque teno virus load at emergency department visit predicts intensive care unit admission of SARS-CoV-2 infected patients.

Accurate prediction of COVID-19 severity remains a challenge. Torque teno virus (TTV), recognized as a surrogate marker of functional immunity in solid organ transplant recipients, holds the potential for assessing infection outcomes. We investigated whether quantifying TTV in nasopharyngeal samples upon emergency department (ED) admission could serve as an early predictor of COVID-19 severity. Retrospective single-center study in the ED of Saint-Louis Hospital in Paris, France. TTV DNA was quantified in nasopharyngeal swab samples collected for SARS-CoV-2 testing. Among 295 SARS-CoV-2 infected patients, 92 returned home, 160 were admitted to medical wards, and 43 to the intensive care unit (ICU). Elevated TTV loads were observed in ICU patients (median: 3.02 log copies/mL, interquartile range [IQR]: 2.215-3.825), exceeding those in discharged (2.215, [0; 2.962]) or hospitalized patients (2.24, [0; 3.29]) (p = 0.006). Multivariate analysis identified diabetes, obesity, hepatitis, fever, dyspnea, oxygen requirement, and TTV load as predictors of ICU admission. A 2.91 log10 copies/mL TTV threshold independently predicted ICU admission. Nasopharyngeal TTV quantification in SARS-CoV-2 infected patients is linked to the likelihood of ICU admission and might reflect respiratory immunosuppression.

Bayesian Sequential Design for Identifying and Ranking Effective Patient Subgroups in Precision Medicine in the Case of Counting Outcome Data with Inflated Zeros.

Precision medicine is revolutionizing health care, particularly by addressing patient variability due to different biological profiles. As traditional treatments may not always be appropriate for certain patient subsets, the rise of biomarker-stratified clinical trials has driven the need for innovative methods. We introduced a Bayesian sequential scheme to evaluate therapeutic interventions in an intensive care unit setting, focusing on complex endpoints characterized by an excess of zeros and right truncation. By using a zero-inflated truncated Poisson model, we efficiently addressed this data complexity. The posterior distribution of rankings and the surface under the cumulative ranking curve (SUCRA) approach provided a comprehensive ranking of the subgroups studied. Different subsets of subgroups were evaluated depending on the availability of biomarker data. Interim analyses, accounting for early stopping for efficacy, were an integral aspect of our design. The simulation study demonstrated a high proportion of correct identification of the subgroup which is the most predictive of the treatment effect, as well as satisfactory false positive and true positive rates. As the role of personalized medicine grows, especially in the intensive care setting, it is critical to have designs that can manage complicated endpoints and that can control for decision error. Our method seems promising in this challenging context.

How to perform prespecified subgroup analyses when using propensity score methods in the case of imbalanced subgroups.

BACKGROUND: Looking for treatment-by-subset interaction on a right-censored outcome based on observational data using propensity-score (PS) modeling is of interest. However, there are still issues regarding its implementation, notably when the subsets are very imbalanced in terms of prognostic features and treatment prevalence. METHODS: We conducted a simulation study to compare two main PS estimation strategies, performed either once on the whole sample ("across subset") or in each subset separately ("within subsets"). Several PS models and estimands are also investigated. We then illustrated those approaches on the motivating example, namely, evaluating the benefits of facial nerve resection in patients with parotid cancer in contact with the nerve, according to pretreatment facial palsy. RESULTS: Our simulation study demonstrated that both strategies provide close results in terms of bias and variance of the estimated treatment effect, with a slight advantage for the "across subsets" strategy in very small samples, provided that interaction terms between the subset variable and other covariates influencing the choice of treatment are incorporated. PS matching without replacement resulted in biased estimates and should be avoided in the case of very imbalanced subsets. CONCLUSIONS: When assessing heterogeneity in the treatment effect in small samples, the "across subsets" strategy of PS estimation is preferred. Then, either a PS matching with replacement or a weighting method must be used to estimate the average treatment effect in the treated or in the overlap population. In contrast, PS matching without replacement should be avoided in this setting.