Optimising precision and power by machine learning in randomised trials with ordinal and time‐to‐event outcomes with an application to COVID‐19

The rapid finding of effective therapeutics requires efficient use of available resources in clinical trials. Covariate adjustment can yield statistical estimates with improved precision, resulting in a reduction in the number of participants required to draw futility or efficacy conclusions. We focus on time‐to‐event and ordinal outcomes. When more than a few baseline covariates are available, a key question for covariate adjustment in randomised studies is how to fit a model relating the outcome and the baseline covariates to maximise precision. We present a novel theoretical result establishing conditions for asymptotic normality of a variety of covariate‐adjusted estimators that rely on machine learning (e.g., ℓ1‐regularisation, Random Forests, XGBoost, and Multivariate Adaptive Regression Splines [MARS]), under the assumption that outcome data are missing completely at random. We further present a consistent estimator of the asymptotic variance. Importantly, the conditions do not require the machine learning methods to converge to the true outcome distribution conditional on baseline variables, as long as they converge to some (possibly incorrect) limit. We conducted a simulation study to evaluate the performance of the aforementioned prediction methods in COVID‐19 trials. Our simulation is based on resampling longitudinal data from over 1500 patients hospitalised with COVID‐19 at Weill Cornell Medicine New York Presbyterian Hospital. We found that using ℓ1‐regularisation led to estimators and corresponding hypothesis tests that control type 1 error and are more precise than an unadjusted estimator across all sample sizes tested. We also show that when covariates are not prognostic of the outcome, ℓ1‐regularisation remains as precise as the unadjusted estimator, even at small sample sizes (n=100). We give an R package adjrct that performs model‐robust covariate adjustment for ordinal and time‐to‐event outcomes.

Optimising precision and power by machine learning in randomised trials with ordinal and time-to-event outcomes with an application to COVID-19.

The rapid finding of effective therapeutics requires efficient use of available resources in clinical trials. Covariate adjustment can yield statistical estimates with improved precision, resulting in a reduction in the number of participants required to draw futility or efficacy conclusions. We focus on time-to-event and ordinal outcomes. When more than a few baseline covariates are available, a key question for covariate adjustment in randomised studies is how to fit a model relating the outcome and the baseline covariates to maximise precision. We present a novel theoretical result establishing conditions for asymptotic normality of a variety of covariate-adjusted estimators that rely on machine learning (e.g., ℓ 1 -regularisation, Random Forests, XGBoost, and Multivariate Adaptive Regression Splines [MARS]), under the assumption that outcome data are missing completely at random. We further present a consistent estimator of the asymptotic variance. Importantly, the conditions do not require the machine learning methods to converge to the true outcome distribution conditional on baseline variables, as long as they converge to some (possibly incorrect) limit. We conducted a simulation study to evaluate the performance of the aforementioned prediction methods in COVID-19 trials. Our simulation is based on resampling longitudinal data from over 1500 patients hospitalised with COVID-19 at Weill Cornell Medicine New York Presbyterian Hospital. We found that using ℓ 1 -regularisation led to estimators and corresponding hypothesis tests that control type 1 error and are more precise than an unadjusted estimator across all sample sizes tested. We also show that when covariates are not prognostic of the outcome, ℓ 1 -regularisation remains as precise as the unadjusted estimator, even at small sample sizes ( n = 100 ). We give an R package adjrct that performs model-robust covariate adjustment for ordinal and time-to-event outcomes.

Comparison of a Target Trial Emulation Framework vs Cox Regression to Estimate the Association of Corticosteroids With COVID-19 Mortality.

Importance: Communication and adoption of modern study design and analytical techniques is of high importance for the improvement of clinical research from observational data. Objective: To compare a modern method for statistical inference, including a target trial emulation framework and doubly robust estimation, with approaches common in the clinical literature, such as Cox proportional hazards models. Design, Setting, and Participants: This retrospective cohort study used longitudinal electronic health record data for outcomes at 28-days from time of hospitalization within a multicenter New York, New York, hospital system. Participants included adult patients hospitalized between March 1 and May 15, 2020, with COVID-19 and not receiving corticosteroids for chronic use. Data were analyzed from October 2021 to March 2022. Exposures: Corticosteroid exposure was defined as more than 0.5 mg/kg methylprednisolone equivalent in a 24-hour period. For target trial emulation, exposures were corticosteroids for 6 days if and when a patient met criteria for severe hypoxia vs no corticosteroids. For approaches common in clinical literature, treatment definitions used for variables in Cox regression models varied by study design (no time frame, 1 day, and 5 days from time of severe hypoxia). Main Outcomes and Measures: The main outcome was 28-day mortality from time of hospitalization. The association of corticosteroids with mortality for patients with moderate to severe COVID-19 was assessed using the World Health Organization (WHO) meta-analysis of corticosteroid randomized clinical trials as a benchmark. Results: A total of 3298 patients (median [IQR] age, 65 [53-77] years; 1970 [60%] men) were assessed, including 423 patients who received corticosteroids at any point during hospitalization and 699 patients who died within 28 days of hospitalization. Target trial emulation analysis found corticosteroids were associated with a reduced 28-day mortality rate, from 32.2%; (95% CI, 30.9%-33.5%) to 25.7% (95% CI, 24.5%-26.9%). This estimate is qualitatively identical to the WHO meta-analysis odds ratio of 0.66 (95% CI, 0.53-0.82). Hazard ratios using methods comparable with current corticosteroid research range in size and direction, from 0.50 (95% CI, 0.41-0.62) to 1.08 (95% CI, 0.80-1.47). Conclusions and Relevance: These findings suggest that clinical research based on observational data can be used to estimate findings similar to those from randomized clinical trials; however, the correctness of these estimates requires designing the study and analyzing the data based on principles that are different from the current standard in clinical research.

Baseline haemoglobin A1c and the risk of COVID-19 hospitalization among patients with diabetes in the INSIGHT Clinical Research Network.

AIMS: To examine the association between baseline glucose control and risk of COVID-19 hospitalization and in-hospital death among patients with diabetes. METHODS: We performed a retrospective cohort study of adult patients in the INSIGHT Clinical Research Network with a diabetes diagnosis and haemoglobin A1c (HbA1c) measurement in the year prior to an index date of March 15, 2020. Patients were divided into four exposure groups based on their most recent HbA1c measurement (in mmol/mol): 39-46 (5.7%-6.4%), 48-57 (6.5%-7.4%), 58-85 (7.5%-9.9%), and ≥86 (10%). Time to COVID-19 hospitalization was compared in the four groups in a propensity score-weighted Cox proportional hazards model adjusting for potential confounders. Patients were followed until June 15, 2020. In-hospital death was examined as a secondary outcome. RESULTS: Of 168,803 patients who met inclusion criteria; 50,016 patients had baseline HbA1c 39-46 (5.7%-6.4%); 54,729 had HbA1c 48-57 (6.5-7.4%); 47,640 had HbA1c 58-85 (7.5^%-9.9%) and 16,418 had HbA1c ≥86 (10%). Compared with patients with HbA1c 48-57 (6.5%-7.4%), the risk of hospitalization was incrementally greater for those with HbA1c 58-85 (7.5%-9.9%) (adjusted hazard ratio [aHR] 1.19, 95% confidence interval [CI] 1.06-1.34) and HbA1c ≥86 (10%) (aHR 1.40, 95% CI 1.19-1.64). The risk of COVID-19 in-hospital death was increased only in patients with HbA1c 58-85 (7.5%-9.9%) (aHR 1.29, 95% CI 1.06, 1.61). CONCLUSIONS: Diabetes patients with high baseline HbA1c had a greater risk of COVID-19 hospitalization, although association between HbA1c and in-hospital death was less consistent. Preventive efforts for COVID-19 should be focused on diabetes patients with poor glucose control.

The Finding My Way UK Clinical Trial: Adaptation Report and Protocol for a Replication Randomized Controlled Efficacy Trial of a Web-Based Psychological Program to Support Cancer Survivors.

BACKGROUND: Cancer survivors frequently report a range of unmet psychological and supportive care needs; these often continue after treatment has finished and are predictive of psychological distress and poor health-related quality of life. Web-based interventions demonstrate good efficacy in addressing these concerns and are more accessible than face-to-face interventions. Finding My Way (FMW) is a web-based, psycho-educational, and cognitive behavioral therapy intervention for cancer survivors developed in Australia. Previous trials have demonstrated that FMW is acceptable, highly adhered to, and effective in reducing the impact of distress on quality of life while leading to cost savings through health resource use reduction. OBJECTIVE: This study aims to adapt the Australian FMW website for a UK cancer care context and then undertake a single-blinded, randomized controlled trial of FMW UK against a treatment-as-usual waitlist control. METHODS: To an extent, our trial design replicates the existing Australian randomized controlled trial of FMW. Following a comprehensive adaptation of the web resource, we will recruit 294 participants (147 per study arm) from across clinical sites in North West England and North Wales. Participants will have been diagnosed with cancer of any type in the last 6 months, have received anticancer treatment with curative intent, be aged ≥16 years, be proficient in English, and have access to the internet and an active email address. Participants will be identified and recruited through the National Institute for Health Research clinical research network. Measures of distress, quality of life, and health economic outcomes will be collected using a self-report web-based questionnaire at baseline, midtreatment, posttreatment, and both 3- and 6-month follow-up. Quantitative data will be analyzed using intention-to-treat mixed model repeated measures analysis. Embedded semistructured qualitative interviews will probe engagement with, and experiences of using, FMW UK and suggestions for future improvements. RESULTS: The website adaptation work was completed in January 2021. A panel of cancer survivors and health care professionals provided feedback on the test version of FMW UK. Feedback was positive overall, although minor updates were made to website navigation, inclusivity, terminology, and the wording of the Improving Communication and Sexuality and Intimacy content. Recruitment for the clinical trial commenced in April 2021. We aim to report on findings from mid-2023. CONCLUSIONS: Replication studies are an important aspect of the scientific process, particularly in psychological and clinical trial literature, especially in different geographical settings. Before replicating the FMW trial in the UK setting, content updating was required. If FMW UK now replicates Australian findings, we will have identified a novel and cost-effective method of psychosocial care delivery for cancer survivors in the United Kingdom. TRIAL REGISTRATION: International Standard Randomized Controlled Trial Number (ISRCTN) 14317248; https://www.isrctn.com/ISRCTN14317248. INTERNATIONAL REGISTERED REPORT IDENTIFIER (IRRID): DERR1-10.2196/31976.

Specialist palliative and end-of-life care for patients with cancer and SARS-CoV-2 infection: a European perspective.

BACKGROUND: Specialist palliative care team (SPCT) involvement has been shown to improve symptom control and end-of-life care for patients with cancer, but little is known as to how these have been impacted by the COVID-19 pandemic. Here, we report SPCT involvement during the first wave of the pandemic and compare outcomes for patients with cancer who received and did not receive SPCT input from multiple European cancer centres. METHODS: From the OnCovid repository (N = 1318), we analysed cancer patients aged ⩾18 diagnosed with COVID-19 between 26 February and 22 June 2020 who had complete specialist palliative care team data (SPCT+ referred; SPCT- not referred). RESULTS: Of 555 eligible patients, 317 were male (57.1%), with a median age of 70 years (IQR 20). At COVID-19 diagnosis, 44.7% were on anti-cancer therapy and 53.3% had ⩾1 co-morbidity. Two hundred and six patients received SPCT input for symptom control (80.1%), psychological support (54.4%) and/or advance care planning (51%). SPCT+ patients had more 'Do not attempt cardio-pulmonary resuscitation' orders completed prior to (12.6% versus 3.7%) and during admission (50% versus 22.1%, p < 0.001), with more SPCT+ patients deemed suitable for treatment escalation (50% versus 22.1%, p < 0.001). SPCT involvement was associated with higher discharge rates from hospital for end-of-life care (9.7% versus 0%, p < 0.001). End-of-life anticipatory prescribing was higher in SPCT+ patients, with opioids (96.3% versus 47.1%) and benzodiazepines (82.9% versus 41.2%) being used frequently for symptom control. CONCLUSION: SPCT referral facilitated symptom control, emergency care and discharge planning, as well as high rates of referral for psychological support than previously reported. Our study highlighted the critical need of SPCTs for patients with cancer during the pandemic and should inform service planning for this population.

Evaluating the impact of COVID-19 on supportive care needs, psychological distress and quality of life in UK cancer survivors and their support network.

OBJECTIVES: The COVID-19 pandemic is having considerable impact on cancer care, including restricted access to hospital-based care, treatment and psychosocial support. We investigated the impact on unmet needs and psychosocial well-being. METHODS: One hundred and forty four participants (77% female), including people with cancer and their support networks, were recruited. The most prevalent diagnosis was breast cancer. Forty-one participants recruited pre-pandemic were compared with 103 participants recruited during the COVID-19 pandemic. We measured participants' unmet supportive care needs, psychological distress and quality of life. RESULTS: Half of our patient respondents reported unexpected changes to treatment following pandemic onset, with widespread confusion about their longer-term consequences. Although overall need levels have not increased, specific needs have changed in prominence. People with cancer reported significantly reduced anxiety (p = 0.049) and improved quality of life (p = 0.032) following pandemic onset, but support network participants reported reduced quality of life (p = 0.009), and non-significantly elevated anxiety, stress and depression. CONCLUSION: Psychological well-being of people with cancer has not been detrimentally affected by pandemic onset. Reliance on home-based support to compensate for the lost availability of structured healthcare pathways may, however, explain significant and detrimental effects on the well-being and quality of life of people in their support and informal care networks.