No full-text available
To read the full-text of this research,
you can request a copy directly from the authors.
Predicting 30-day mortality with routine blood tests in patients undergoing palliative radiation therapy: A comparison of logistic regression and gradient boosting models
ResearchGate has not been able to resolve any citations for this publication.
Background
The role of neutrophil–lymphocyte ratio (NLR) as a predictor for survival in single fraction SBRT-treated non-small cell lung cancer (NSCLC) patients remains unclear. We performed an observational cohort study to determine the role of pretreatment NLR in predicting survival of early-stage NSCLC patients after single fraction SBRT.
Methods
A single-institution database of peripheral early-stage NSCLC patients treated with SBRT from February 2007 to May 2022 was queried. Optimal threshold of neutrophil–lymphocyte ratio (NLR) was defined based on maximally selected rank statistics. Cox multivariable analysis (MVA), Kaplan–Meier, and propensity score matching were performed to evaluate outcomes.
Results
A total of 286 patients were included for analysis with median follow up of 19.7 months. On Cox multivariate analysis, as a continuous variable, NLR was shown to be an independent predictor of OS (adjusted hazards ratio [aHR] 1.06, 95% CI 1.02–1.10, p = 0.005) and PFS (aHR 1.05, 95% CI 1.01–1.09, p = 0.013). In addition, NLR was associated with DF (aHR 1.11, 95% CI 1.05–1.18, p < 0.001). Maximally selected rank statistics determined 3.28 as the cutoff point of high NLR versus low NLR. These findings were confirmed upon propensity matching.
Conclusions
Pretreatment NLR is an independent predictor for survival outcomes of peripheral early-stage NSCLC patients after single fraction SBRT.
Simple Summary
To investigate the adoption of machine learning in palliative care research and clinical practice, we systematically searched for published research papers on the topic. We found several publications that used different kinds of machine learning in palliative care for different use cases. However, on average, there needs to be more rigorous testing of the models to ensure that they work well in different settings.
Abstract
Objective: To summarize the available literature on using machine learning (ML) for palliative care practice as well as research and to assess the adherence of the published studies to the most important ML best practices. Methods: The MEDLINE database was searched for the use of ML in palliative care practice or research, and the records were screened according to PRISMA guidelines. Results: In total, 22 publications using machine learning for mortality prediction (n = 15), data annotation (n = 5), predicting morbidity under palliative therapy (n = 1), and predicting response to palliative therapy (n = 1) were included. Publications used a variety of supervised or unsupervised models, but mostly tree-based classifiers and neural networks. Two publications had code uploaded to a public repository, and one publication uploaded the dataset. Conclusions: Machine learning in palliative care is mainly used to predict mortality. Similarly to other applications of ML, external test sets and prospective validations are the exception.
PURPOSE
Patients with advanced solid tumors may receive intensive treatments near the end of life. This study aimed to create a machine learning (ML) model using limited features to predict 6-month mortality at treatment decision points (TDPs).
METHODS
We identified a cohort of adults with advanced solid tumors receiving care at a major cancer center from 2014 to 2020. We identified TDPs for new lines of therapy (LoTs) and confirmed mortality at 6 months after a TDP. Using extreme gradient boosting, ML models were developed, which used or derived features from a limited set of electronic health record data considering the literature, clinical relevance, variability, availability, and predictive importance using Shapley additive explanations scores. We predicted and observed 6-month mortality after a TDP and assessed a risk stratification strategy with different risk thresholds to support communication of chance of survival.
RESULTS
Four thousand one hundred ninety-two patients were included. Patients had 7,056 TDPs, for which the 6-month mortality increased from 17.9% to 46.7% after starting first to sixth LoT, respectively. On the basis of internal validation, models using both 111 (Full) or 45 (Limited-45) features accurately predicted 6-month mortality (area under the curve ≥ 0.80). Using a 0.3 risk threshold in the Limited-45 model, the observed 6-month survival was 34% (95% CI, 28 to 40) versus 81% (95% CI, 81 to 82) among those classified with low or higher chance of survival, respectively. The positive predictive value of the Limited-45 model was 0.66 (95% CI, 0.60 to 0.72).
CONCLUSION
We developed and validated a ML model using a limited set of 45 features readily derived from electronic health record data to predict 6-month prognosis in patients with advanced solid tumors. The model output may support shared decision making as patients consider the next LoT.
Background
The expected 30-day mortality rate for patients treated with palliative radiation is not established. The primary objective of this study is to define the proportion of patients with advanced cancer who die within 30-days of palliative radiotherapy (PR). Additionally, we explored the short term survival of patient subgroups undergoing PR treatment.
Methods
We searched MEDLINE, CINAHL, Embase and Cochrane Database of Systematic Reviews from January 1st 1980 to June 26, 2020. We included PUBMED’s “related search” and reference lists to further identify of articles. A meta-analysis of these research studies and reviews was performed. Published and unpublished English language randomized controlled trials, observational or prospective studies, and systematic reviews that reported 30-day mortality for patients with advanced cancer who received PR were eligible. Data extraction was done by two independent authors and included study quality indicators. To improve distribution and variance, all proportions were transformed using logit transformation. A random-effects model was used to pool data, using Der Simonian and Laird method of estimation where possible and appropriate.
Results
The data from 42 studies contributing 88,516 patients with advanced cancer who received PR were evaluated. The summary proportion of mortality in patients with advanced cancer within 30 days of receiving PR was 16% (95% CI = 14% to 18%). We found substantial heterogeneity in our data (I² = 98.76%, p < 0.001), hence we applied subgroup analysis to identify potential moderating factors. We found a higher 30-day mortality rate after PR in the following groups: multiple treatment sites (QM(1)=9.54, p=0.002), hepatobiliary primary (QM(1)=24.20, p<0.001), inpatient status (QM(1)=92.27, p<0.001), Eastern Cooperative Oncology Group performance status (ECOG) 3-4 (QM(1)=8.70, p=0.003), United States (U.S.) patients (QM(1)=28.70, p<0.001) among others.
Conclusions
We found that 16% of patients with advanced cancer receiving PR die within 30 days of treatment. Our finding can be used as a benchmark to establish a global quality metric for radiation oncology practice audits.
Background and aim:
The prognostic assessment of patients referred for palliative radiotherapy can be conducted by site-specific scores. A quick assessment that would cover the whole spectrum could simplify the working day of clinicians who are not specialists for a particular disease site. This study evaluated a promising score, the LabBM (validated for brain metastases), in patients treated for other indications.
Materials and methods:
The LabBM score was calculated in 375 patients by assigning 1 point each for C-reactive protein and lactate dehydrogenase above the upper limit of normal, and 0.5 points each for hemoglobin, platelets and albumin below the lower limit of normal. Uni- and multivariate analyses were performed.
Results:
Median overall survival gradually decreased with increasing point sum (range 25.1-1.1 months). When grouped according to the original three-tiered model, excellent discrimination was found. Patients with 0-1 points had a median survival of 15.7 months. Those with 1.5-2 points had a median survival of 5.8 months. Finally, those with 2.5-3.5 points had a median survival of 3.2 months (all p-values ≤ 0.001).
Conclusion:
The LabBM score, which is derived from inexpensive blood tests and easy to use, stratified patients into three very distinct prognostic groups and deserves further validation.
PurposeThe purpose of this study was to develop a simple prognostic model based on objective indicators alone, i.e., routine blood test data, without using any subjective variables such as patient’s symptoms and physician’s prediction.Methods
The subjects of this retrospective study were patients at the palliative care unit of Tohoku University Hospital, Japan. Eligible patients were over 20 years old and had advanced cancer (n = 225). The model for predicting survival was developed based on Cox proportional hazards regression models for univariable and multivariable analyses of 20 items selected from routine blood test data. All the analyses were performed according to the TRIPOD statement (https://www.tripod-statement.org/).ResultsThe univariable and multivariable regression analyses identified total bilirubin, creatinine, urea/creatinine ratio, aspartate aminotransferase, albumin, total leukocyte count, differential lymphocyte count, and platelet/lymphocyte ratio as significant risk factors for mortality. Based on the hazard ratios, the area under the curve for the new risk model was 0.87 for accuracy, 0.83 for sensitivity, and 0.74 for specificity. Diagnostic accuracy was higher than provided by the Palliative Prognostic Score and the Palliative Prognostic Index. The Kaplan–Meier analysis demonstrated a survival significance of classifying patients according to their score into low-, medium-, and high-mortality risk groups having median survival times of 67 days, 34 days, and 11 days, respectively (p < 0.001).Conclusions
We developed a simple and accurate prognostic model for predicting the survival of patients with advanced cancer based on routine blood test values alone that may be useful for appropriate advanced care planning in a palliative care setting.
Importance
Serious illness conversations (SICs) are structured conversations between clinicians and patients about prognosis, treatment goals, and end-of-life preferences. Interventions that increase the rate of SICs between oncology clinicians and patients may improve goal-concordant care and patient outcomes.
Objective
To determine the effect of a clinician-directed intervention integrating machine learning mortality predictions with behavioral nudges on motivating clinician-patient SICs.
Design, Setting, and Participants
This stepped-wedge cluster randomized clinical trial was conducted across 20 weeks (from June 17 to November 1, 2019) at 9 medical oncology clinics (8 subspecialty oncology and 1 general oncology clinics) within a large academic health system in Pennsylvania. Clinicians at the 2 smallest subspecialty clinics were grouped together, resulting in 8 clinic groups randomly assigned to the 4 intervention wedge periods. Included participants in the intention-to-treat analyses were 78 oncology clinicians who received SIC training and their patients (N = 14 607) who had an outpatient oncology encounter during the study period.
Interventions
(1) Weekly emails to oncology clinicians with SIC performance feedback and peer comparisons; (2) a list of up to 6 high-risk patients (≥10% predicted risk of 180-day mortality) scheduled for the next week, estimated using a validated machine learning algorithm; and (3) opt-out text message prompts to clinicians on the patient’s appointment day to consider an SIC. Clinicians in the control group received usual care consisting of weekly emails with cumulative SIC performance.
Main Outcomes and Measures
Percentage of patient encounters with an SIC in the intervention group vs the usual care (control) group.
Results
The sample consisted of 78 clinicians and 14 607 patients. The mean (SD) age of patients was 61.9 (14.2) years, 53.7% were female, and 70.4% were White. For all encounters, SICs were conducted among 1.3% in the control group and 4.6% in the intervention group, a significant difference (adjusted difference in percentage points, 3.3; 95% CI, 2.3-4.5; P < .001). Among 4124 high-risk patient encounters, SICs were conducted among 3.6% in the control group and 15.2% in the intervention group, a significant difference (adjusted difference in percentage points, 11.6; 95% CI, 8.2-12.5; P < .001).
Conclusions and Relevance
In this stepped-wedge cluster randomized clinical trial, an intervention that delivered machine learning mortality predictions with behavioral nudges to oncology clinicians significantly increased the rate of SICs among all patients and among patients with high mortality risk who were targeted by the intervention. Behavioral nudges combined with machine learning mortality predictions can positively influence clinician behavior and may be applied more broadly to improve care near the end of life.
Trial Registration
ClinicalTrials.gov Identifier: NCT03984773
We developed a predictive score system for 30-day mortality after palliative radiotherapy by using predictors from routine electronic medical record. Patients with metastatic cancer receiving first course palliative radiotherapy from 1 July, 2007 to 31 December, 2017 were identified. 30-day mortality odds ratios and probabilities of the death predictive score were obtained using multivariable logistic regression model. Overall, 5,795 patients participated. Median follow-up was 39.6 months (range, 24.5–69.3) for all surviving patients. 5,290 patients died over a median 110 days, of whom 995 (17.2%) died within 30 days of radiotherapy commencement. The most important mortality predictors were primary lung cancer (odds ratio: 1.73, 95% confidence interval: 1.47–2.04) and log peripheral blood neutrophil lymphocyte ratio (odds ratio: 1.71, 95% confidence interval: 1.52–1.92). The developed predictive scoring system had 10 predictor variables and 20 points. The cross-validated area under curve was 0.81 (95% confidence interval: 0.79–0.82). The calibration suggested a reasonably good fit for the model (likelihood-ratio statistic: 2.81, P = 0.094), providing an accurate prediction for almost all 30-day mortality probabilities. The predictive scoring system accurately predicted 30-day mortality among patients with stage IV cancer. Oncologists may use this to tailor palliative therapy for patients.
Conventional prognostic scores for terminally ill cancer patients may have less objectivity because they include subjective or categorical variables that do not consider intensity or severity. The aim of this study was to identify prognostic factors for 30-day mortality from routine blood examination of terminally ill cancer patients. A total of 1308 study patients in a hospice setting were divided into investigation (n=761) and validation (n=547) groups. Twenty laboratory blood parameters were analyzed. Multivariate analysis revealed that ten variables (C-reactive protein ≥5.4 mg/dL, serum albumin <2.8 g/dL, blood urea nitrogen ≥21 mg/dL, white blood cell count ≥8.600 × 10³/μL, eosinophil percentage <0.8%, neutrophil-to-lymphocyte ratio ≥11.1, hemoglobin level ≥ 13.2 g/dL, mean corpuscular volume ≥ 93.7 fl, red cell distribution width ≥ 16, and platelet count < 159 × 10³/μL) were significant independent prognostic factors for 30-day survival. The laboratory prognostic score (LPS) was calculated by the sum of blood indices among the ten variables. The LPS showed acceptable accuracy for 30-day mortality in the investigation and validation groups. LPS 5 (including any five factors) predicted death within 30 days, with a sensitivity of 85%, a specificity of 55%, a positive predictive value of 72%, and a negative predictive value of 74%. The predictive value of LPS was comparable to those of conventional prognostic scores, which include signs and symptoms. The LPS can provide additional information to conventional prognostic scores.
Importance
Patients with cancer who die soon after starting chemotherapy incur costs of treatment without the benefits. Accurately predicting mortality risk before administering chemotherapy is important, but few patient data–driven tools exist.
Objective
To create and validate a machine learning model that predicts mortality in a general oncology cohort starting new chemotherapy, using only data available before the first day of treatment.
Design, Setting, and Participants
This retrospective cohort study of patients at a large academic cancer center from January 1, 2004, through December 31, 2014, determined date of death by linkage to Social Security data. The model was derived using data from 2004 through 2011, and performance was measured on nonoverlapping data from 2012 through 2014. The analysis was conducted from June 1 through August 1, 2017. Participants included 26 946 patients starting 51 774 new chemotherapy regimens.
Main Outcomes and Measures
Thirty-day mortality from the first day of a new chemotherapy regimen. Secondary outcomes included model discrimination by predicted mortality risk decile among patients receiving palliative chemotherapy, and 180-day mortality from the first day of a new chemotherapy regimen.
Results
Among the 26 946 patients included in the analysis, mean age was 58.7 years (95% CI, 58.5-58.9 years); 61.1% were female (95% CI, 60.4%-61.9%); and 86.9% were white (95% CI, 86.4%-87.4%). Thirty-day mortality from chemotherapy start was 2.1% (95% CI, 1.9%-2.4%). Among the 9114 patients in the validation set, the most common primary cancers were breast (21.1%; 95% CI, 20.2%-21.9%), colorectal (19.3%; 95% CI, 18.5%-20.2%), and lung (18.0%; 95% CI, 17.2%-18.8%). Model predictions were accurate for all patients (area under the curve [AUC], 0.940; 95% CI, 0.930-0.951). Predictions for patients starting palliative chemotherapy (46.6% of regimens; 95% CI, 45.8%-47.3%), for whom prognosis is particularly important, remained highly accurate (AUC, 0.924; 95% CI, 0.910-0.939). To illustrate model discrimination, patients were ranked initiating palliative chemotherapy by model-predicted mortality risk, and observed mortality was calculated by risk decile. Thirty-day mortality in the highest-risk decile was 22.6% (95% CI, 19.6%-25.6%); in the lowest-risk decile, no patients died. Predictions remained accurate across all primary cancers, stages, and chemotherapies, even for clinical trial regimens that first appeared in years after the model was trained (AUC, 0.942; 95% CI, 0.882-1.000). The same model also performed well for prediction of 180-day mortality (AUC for all patients, 0.870 [95% CI, 0.862-0.877]; highest- vs lowest-risk decile mortality, 74.8% [95% CI, 72.7%-77.0%] vs 0.2% [95% CI, 0.01%-0.4%]). Predictions were more accurate than estimates from randomized clinical trials of individual chemotherapies or the Surveillance, Epidemiology, and End Results data set.
Conclusions and Relevance
A machine learning algorithm using electronic health record data accurately predicted short-term mortality among patients starting chemotherapy. Further research is necessary to determine the generalizability and feasibility of applying this algorithm in clinical settings.
Understanding why a model makes a certain prediction can be as crucial as the prediction's accuracy in many applications. However, the highest accuracy for large modern datasets is often achieved by complex models that even experts struggle to interpret, such as ensemble or deep learning models, creating a tension between accuracy and interpretability. In response, various methods have recently been proposed to help users interpret the predictions of complex models, but it is often unclear how these methods are related and when one method is preferable over another. To address this problem, we present a unified framework for interpreting predictions, SHAP (SHapley Additive exPlanations). SHAP assigns each feature an importance value for a particular prediction. Its novel components include: (1) the identification of a new class of additive feature importance measures, and (2) theoretical results showing there is a unique solution in this class with a set of desirable properties. The new class unifies six existing methods, notable because several recent methods in the class lack the proposed desirable properties. Based on insights from this unification, we present new methods that show improved computational performance and/or better consistency with human intuition than previous approaches.
Gradient boosting machines are a family of powerful machine-learning techniques that have shown considerable success in a wide range of practical applications. They are highly customizable to the particular needs of the application, like being learned with respect to different loss functions. This article gives a tutorial introduction into the methodology of gradient boosting methods with a strong focus on machine learning aspects of modeling. A theoretical information is complemented with descriptive examples and illustrations which cover all the stages of the gradient boosting model design. Considerations on handling the model complexity are discussed. Three practical examples of gradient boosting applications are presented and comprehensively analyzed.
We performed a Monte Carlo study to evaluate the effect of the number of events per variable (EPV) analyzed in logistic regression analysis. The simulations were based on data from a cardiac trial of 673 patients in which 252 deaths occurred and seven variables were cogent predictors of mortality; the number of events per predictive variable was (252/7 =) 36 for the full sample. For the simulations, at values of EPV = 2, 5, 10, 15, 20, and 25, we randomly generated 500 samples of the 673 patients, chosen with replacement, according to a logistic model derived from the full sample. Simulation results for the regression coefficients for each variable in each group of 500 samples were compared for bias, precision, and significance testing against the results of the model fitted to the original sample. For EPV values of 10 or greater, no major problems occurred. For EPV values less than 10, however, the regression coefficients were biased in both positive and negative directions; the large sample variance estimates from the logistic model both overestimated and underestimated the sample variance of the regression coefficients; the 90% confidence limits about the estimated values did not have proper coverage; the Wald statistic was conservative under the null hypothesis; and paradoxical associations (significance in the wrong direction) were increased. Although other factors (such as the total number of events, or sample size) may influence the validity of the logistic model, our findings indicate that low EPV can lead to major problems.
Function approximation is viewed from the perspective of numerical optimization in function space, rather than parameter space. A connection is made between stagewise additive expansions and steepest--descent minimization. A general gradient--descent "boosting" paradigm is developed for additive expansions based on any fitting criterion. Specific algorithms are presented for least--squares, least--absolute--deviation, and Huber--M loss functions for regression, and multi--class logistic likelihood for classification. Special enhancements are derived for the particular case where the individual additive components are decision trees, and tools for interpreting such "TreeBoost" models are presented. Gradient boosting of decision trees produces competitive, highly robust, interpretable procedures for regression and classification, especially appropriate for mining less than clean data. Connections between this approach and the boosting methods of Freund and Shapire 1996, and Fr...
This prognostic study performed external validation of a machine learning model to predict 6-month mortality among patients with advanced solid tumors.
In patients with advanced cancer, radiation therapy is considered at various time points in the patient's clinical course from diagnosis to death. As some patients are living longer with metastatic cancer on novel therapeutics, radiation oncologists are increasingly using radiation therapy as an ablative therapy in appropriately selected patients. However, most patients with metastatic cancer still eventually die of their disease. For those without effective targeted therapy options or those who are not candidates for immunotherapy, the time frame from diagnosis to death is still relatively short. Given this evolving landscape, prognostication has become increasingly challenging. Thus, radiation oncologists must be diligent about defining the goals of therapy and considering all treatment options from ablative radiation to medical management and hospice care. The risks and benefits of radiation therapy vary based on an individual patient's prognosis, goals of care, and the ability of radiation to help with their cancer symptoms without undue toxicity over the course of their expected lifetime. When considering recommending a course of radiation, physicians must broaden their understanding of risks and benefits to include not only physical symptoms, but also various psychosocial burdens. These include financial burdens to the patient, to their caregiver and to the healthcare system. The burden of time spent at the end-of-life receiving radiation therapy must also be considered. Thus, the consideration of radiation therapy at the end-of-life can be complex and requires careful attention to the whole patient and their goals of care.
Estimation of patient prognosis plays a central role in guiding decision making for the palliative management of metastatic disease, and a number of statistical models have been developed to provide survival estimates for patients in this context. In this review, we discuss several well-validated survival prediction models for patients receiving palliative radiotherapy to sites outside of the brain. Key considerations include the type of statistical model, model performance measures and validation procedures, studies' source populations, time points used for prognostication, and details of model output. We then briefly discuss underutilization of these models, the role of decision support aids, and the need to incorporate patient preference in shared decision making for patients with metastatic disease who are candidates for palliative radiotherapy.
PURPOSE
Given the heterogeneity and improvement in outcomes for metastatic breast cancer (MBC), we developed a staging system that refines prognostic estimates for patients with metastatic cancer at the time of initial diagnosis, de novo MBC (dnMBC), on the basis of survival outcomes and disease-related variables.
METHODS
Patients with dnMBC (2010-2016) were selected from the National Cancer Database (NCDB). Recursive partitioning analysis (RPA) was used to group patients with similar overall survival (OS) on the basis of clinical T category, grade, estrogen receptor (ER), progesterone receptor, human epidermal growth factor receptor 2, histology, organ system site of metastases (bone-only, brain-only, visceral), and number of organ systems involved. Three-year OS rates were used to assign a final stage: IVA: >70%, IVB: 50%-70%, IVC: 25 to <50%, and IVD: <25%. Bootstrapping was applied with 1,000 iterations, and final stage assignments were made based on the most commonly occurring assignment. Unadjusted OS was estimated. Validation analyses were conducted using SEER and NCDB.
RESULTS
At a median follow-up of 52.9 months, the median OS of the original cohort (N = 42,467) was 35.4 months (95% CI, 34.8 to 35.9). RPA stratified patients into 53 groups with 3-year OS rates ranging from 73.5% to 5.7%; these groups were amalgamated into four stage groups: 3-year OS, A = 73.2%, B = 61.9%, C = 40.1%, and D = 17% (log-rank P < .001). After bootstrapping, the survival outcomes for the four stages remained significantly different (log-rank P < .001). This staging system was then validated using SEER data (N = 20,469) and a separate cohort from the NCDB (N = 7,645) (both log-rank P < .001).
CONCLUSION
Our findings regarding the heterogeneity in outcomes for patients with dnMBC could guide future revisions of the current American Joint Committee on Cancer staging guidelines for patients with newly diagnosed stage IV disease. Our findings should be independently confirmed.
The treatment paradigm for patients with metastatic cancer has shifted in the era of oligometastatic disease,¹ novel systemic therapy, and local ablative therapy options such as stereotactic ablative body radiotherapy (SABR). The equation of “metastatic” with “incurable” has been challenged, as an increasing number of patients with stage IV disease have long-term survival outcomes. In the treatment of oligometastasis, several clinical trials have recently demonstrated the survival benefits of SABR.²⁻⁴ For patients with diffuse metastatic disease or those who are otherwise being treated with palliative intent, the optimal radiation technique has also been a subject of scrutiny.⁵ To this end, in 2016 the Department of Radiation Oncology at Memorial Sloan Kettering Cancer Center established a dedicated program for patients with metastatic cancer to address the specific needs of these patients, the Precision Radiation for Oligometastatic and Metastatic Disease, or PROMISE, Program.
Background
To determine if a machine learning approach optimizes survival estimation for patients with symptomatic bone metastases (SBM), we developed the Bone Metastases Ensemble Trees for Survival (BMETS) to predict survival using 27 prognostic covariates. To establish relative clinical utility, we compared BMETS to two simpler Cox regression models used in this setting.
Methods and Materials
For 492 bone sites in 397 patients evaluated for palliative radiation therapy (RT) for SBM from 1/2007-1/2013, data for 27 clinical variables were collected. These covariates and the primary outcome of time from consultation to death were used to build BMETS utilizing random survival forests. We then performed Cox regressions as per two validated models: Chow’s 3-item (C-3) and Westhoff’s 2-item (W-2) tools. Model performance was assessed using cross-validation procedures and measured by time-dependent area under the curve (tAUC) for all three models. For temporal validation, a separate dataset comprised of 104 bone sites treated in 85 patients in 2018 was used to estimate tAUC from BMETS.
Results
Median survival was 6.3 months. Variable importance was greatest for performance status, blood cell counts, recent systemic therapy type, and receipt of concurrent non-bone palliative RT. tAUC at 3-, 6-, and 12-months was 0.83, 0.81, and 0.81, respectively, suggesting excellent discrimination of BMETS across post-consultation time points. BMETS outperformed simpler models at each time, with respective tAUC at each time of 0.78, 0.76, and 0.74 for the C-3 model and 0.80, 0.78, and 0.77 for the W-2 model. For the temporal validation set, respective tAUC was similarly high at 0.86, 0.82, and 0.78.
Conclusions
For patients with SBM, BMETS improved survival predictions versus simpler traditional models. Model performance was maintained when applied to a temporal validation set. To facilitate clinical use, we developed a web platform for data entry and display of BMETS predicted survival probabilities.
Background:
The oligometastatic paradigm suggests that some patients with a limited number of metastases might be cured if all lesions are eradicated. Evidence from randomised controlled trials to support this paradigm is scarce. We aimed to assess the effect of stereotactic ablative radiotherapy (SABR) on survival, oncological outcomes, toxicity, and quality of life in patients with a controlled primary tumour and one to five oligometastatic lesions.
Methods:
This randomised, open-label phase 2 study was done at 10 hospitals in Canada, the Netherlands, Scotland, and Australia. Patients aged 18 or older with a controlled primary tumour and one to five metastatic lesions, Eastern Cooperative Oncology Group score of 0-1, and a life expectancy of at least 6 months were eligible. After stratifying by the number of metastases (1-3 vs 4-5), we randomly assigned patients (1:2) to receive either palliative standard of care treatments alone (control group), or standard of care plus SABR to all metastatic lesions (SABR group), using a computer-generated randomisation list with permuted blocks of nine. Neither patients nor physicians were masked to treatment allocation. The primary endpoint was overall survival. We used a randomised phase 2 screening design with a two-sided α of 0·20 (wherein p<0·20 designates a positive trial). All analyses were intention to treat. This study is registered with ClinicalTrials.gov, number NCT01446744.
Findings:
99 patients were randomised between Feb 10, 2012, and Aug 30, 2016. Of 99 patients, 33 (33%) were assigned to the control group and 66 (67%) to the SABR group. Two (3%) patients in the SABR group did not receive allocated treatment and withdrew from the trial; two (6%) patients in the control group also withdrew from the trial. Median follow-up was 25 months (IQR 19-54) in the control group versus 26 months (23-37) in the SABR group. Median overall survival was 28 months (95% CI 19-33) in the control group versus 41 months (26-not reached) in the SABR group (hazard ratio 0·57, 95% CI 0·30-1·10; p=0·090). Adverse events of grade 2 or worse occurred in three (9%) of 33 controls and 19 (29%) of 66 patients in the SABR group (p=0·026), an absolute increase of 20% (95% CI 5-34). Treatment-related deaths occurred in three (4·5%) of 66 patients after SABR, compared with none in the control group.
Interpretation:
SABR was associated with an improvement in overall survival, meeting the primary endpoint of this trial, but three (4·5%) of 66 patients in the SABR group had treatment-related death. Phase 3 trials are needed to conclusively show an overall survival benefit, and to determine the maximum number of metastatic lesions wherein SABR provides a benefit.
Funding:
Ontario Institute for Cancer Research and London Regional Cancer Program Catalyst Grant.
Background:
Multiple reports have attributed a prognostic value to routine blood tests results for patients with glioblastoma. However, these studies have reported conflicting results and have often had small sample sizes. We sought to validate the prognostic value of the described tests in an independent glioblastoma patient population.
Methods:
We performed a retrospective single-center multivariable analysis of 497 patients with glioblastoma who had postoperatively undergone radiotherapy and/or chemotherapy to identify the prognostic value for median overall survival of hemoglobin, white blood cell, monocyte, neutrophil, leukocyte, and platelet counts, neutrophil/lymphocyte ratio, C-reactive protein, erythrocyte sedimentation rate, activated partial thromboplastin time, prothrombin time, and lactate dehydrogenase. We also evaluated known prognostic factors for survival such as patient age, intervention type, IDH1 status, Karnofsky clinical performance status, and postoperative treatment modality.
Results:
In a multivariable model, after correcting for multiple testing bias, biopsy alone (hazard ratio, 0.35; 95% confidence interval, 0.26-0.49; false discovery rate-adjusted P < 0.001) and monotherapy after surgery (hazard ratio, 0.46; 95% confidence interval, 0.33-0.66; false discovery rate-adjusted P < 0.001) remained significantly associated with worse median overall survival. Patient age and Karnofsky performance status score ≥70 did not significantly influence survival in the multivariable model. No routine blood test included in the multivariable analysis was significantly associated with survival.
Conclusions:
In the present study, hemoglobin, white blood cell, monocyte, neutrophil, leukocyte, and platelet counts, neutrophil/lymphocyte ratio, C-reactive protein, erythrocyte sedimentation rate, activated partial thromboplastin time, prothrombin time, and lactate dehydrogenase levels did not independently predict for overall survival in patients with glioblastoma.
We propose a new method for estimation in linear models. The ‘lasso’ minimizes the residual sum of squares subject to the sum of the absolute value of the coefficients being less than a constant. Because of the nature of this constraint it tends to produce some coefficients that are exactly 0 and hence gives interpretable models. Our simulation studies suggest that the lasso enjoys some of the favourable properties of both subset selection and ridge regression. It produces interpretable models like subset selection and exhibits the stability of ridge regression. There is also an interesting relationship with recent work in adaptive function estimation by Donoho and Johnstone. The lasso idea is quite general and can be applied in a variety of statistical models: extensions to generalized regression models and tree‐based models are briefly described.
Purpose:
With rapidly evolving treatment options in cancer, the complexity in the clinical decision-making process for oncologists represents a growing challenge magnified by oncologists' disposition of intuition-based assessment of treatment risks and overall mortality. Given the unmet need for accurate prognostication with meaningful clinical rationale, we developed a highly interpretable prediction tool to identify patients with high mortality risk before the start of treatment regimens.
Methods:
We obtained electronic health record data between 2004 and 2014 from a large national cancer center and extracted 401 predictors, including demographics, diagnosis, gene mutations, treatment history, comorbidities, resource utilization, vital signs, and laboratory test results. We built an actionable tool using novel developments in modern machine learning to predict 60-, 90- and 180-day mortality from the start of an anticancer regimen. The model was validated in unseen data against benchmark models.
Results:
We identified 23,983 patients who initiated 46,646 anticancer treatment lines, with a median survival of 514 days. Our proposed prediction models achieved significantly higher estimation quality in unseen data (area under the curve, 0.83 to 0.86) compared with benchmark models. We identified key predictors of mortality, such as change in weight and albumin levels. The results are presented in an interactive and interpretable tool ( www.oncomortality.com ).
Conclusion:
Our fully transparent prediction model was able to distinguish with high precision between highest- and lowest-risk patients. Given the rich data available in electronic health records and advances in machine learning methods, this tool can have significant implications for value-based shared decision making at the point of care and personalized goals-of-care management to catalyze practice reforms.
Purpose:
To investigate the utilization of palliative radiation therapy (RT), predictors for the use of RT, and symptom palliation following RT during the last 30 days of life through systemic review of literature.
Materials/methods:
A systematic search of available medical literature databases was performed on patients receiving palliative RT in the last 30 days of life. A total of 18 studies were evaluated.
Results:
The overall palliative RT utilization rates during the last month of life were in the range of 5-10% among patients who died of cancer and 9-15.3% of patients who received palliative RT. The most commonly used regimen was 30 Gy in 10 fractions (36-90%). Single fraction RT utilization ranged from 0% to 59%. ECOG performance status 3-4 was significantly associated with patients receiving RT in the last 30 days of life and shorter survival. Twenty-six percent of patients who survived less than 1 month were reported to show symptom palliation following RT.
Conclusion:
Palliative RT was performed in approximately 10% of patients who died of cancer near their end of life, with the most commonly used regimen of 30 Gy in 10 fractions. This study suggests that greater use of shorter or single fraction regimens may be beneficial, especially in patients with poor performance status.
Objectives/hypothesis:
Poor nutritional status in patients with head and neck squamous cell carcinoma (HNSCC) is associated with tumor progression and survival. This study examined the prognostic value of nutritional and hematological markers in patients with HNSCC who received definitive treatments.
Study design:
A prospective observational cohort study.
Methods:
This study included 338 consecutive patients who underwent surgery and/or radiotherapy/chemoradiotherapy for treatment-naïve HNSCC. Body weight and nutritional and hematological parameters were regularly measured before and after treatment. Univariate and multivariate analyses using Cox proportional hazards models were performed to identify factors associated with disease-free survival (DFS), cancer-specific survival (CSS), and overall survival (OS).
Results:
Body weight, serum total protein and albumin levels, and hematological variables significantly decreased after treatment. Univariate analyses illustrated that age, tumor site, T and N classifications, overall stage, pretreatment serum albumin (<3.5 g/dL) and hemoglobin (<12 g/dL) levels, and neutrophil-lymphocyte ratio were significantly associated with DFS, CSS, and OS (all P < .05). Multivariate analyses identified age, tumor site, N classification, and pretreatment albumin levels as independent predictors of DFS, CSS, and OS (all P < .05). Patients with low serum albumin levels prior to treatment experienced approximately sixfold increases in the risks of tumor progression and cancer-specific and overall mortality compared to the findings in their counterparts.
Conclusions:
Our results suggest that pretreatment serum albumin levels predict DFS, CSS, and OS in patients who received definitive treatment for HNSCC. These findings might help to predict treatment outcome and guide nutritional intervention in patients with HNSCC.
Level of evidence:
2b Laryngoscope, 2017.
Background:
We aimed to investigate the potential of standard hematologic and serum biochemical parameters to provide an independent and substantial contribution to the prediction of survival in patients with newly diagnosed brain metastases (BM).
Methods:
Hemoglobin, white blood cell count, platelet count, serum albumin, creatinine, lactate dehydrogenase (LDH), and C-reactive protein (CRP) were assessed at diagnosis of BM in a discovery cohort of 1200 cancer patients. A multivariable Cox regression model was used to derive the LabBM score. The LabBM score was externally validated in an independent cohort consisting of 366 patients.
Results:
Hemoglobin below lower limit of normal (<LLN; hazard ratio [HR] 1.28; P = .001), platelet count <LLN (HR: 1.36; P = .013), albumin <LLN (HR: 1.19; P = .038), LDH above upper limit of normal (>ULN; HR: 1.51; P < .001), and CRP >ULN (HR: 1.52; P < .001) were associated with survival in a multivariable Cox regression model and were included in the calculation of the LabBM score. Multivariable analysis including the LabBM score and graded prognostic assessment class revealed an independent and significant association of the LabBM score with overall survival (OS) (HR: 1.42; 95% CI: 1.29-1.57; P < .001). The strong and independent association of LabBM score (HR: 1.93; 95% CI: 1.54-2.42) with OS prognosis was confirmed in the validation cohort.
Conclusion:
Standard clinical blood parameters, combined in the easy-to-calculate LabBM score, provide strong and independent prognostic information in patients with BM. The LabBM score is an objective, inexpensive, and reproducible tool to plan clinical management strategies in BM patients and to improve patient selection and stratification for clinical trials.
Tree boosting is a highly effective and widely used machine learning method. In this paper, we describe a scalable end-to-end tree boosting system called XGBoost, which is used widely by data scientists to achieve state-of-the-art results on many machine learning challenges. We propose a novel sparsity-aware algorithm for sparse data and weighted quantile sketch for approximate tree learning. More importantly, we provide insights on cache access patterns, data compression and sharding to build a scalable tree boosting system. By combining these insights, XGBoost scales beyond billions of examples using far fewer resources than existing systems.
Predicting life expectancy (LE) in patients with metastatic cancer who are receiving palliative therapies is a difficult task. The purpose of the current study was to develop a LE prediction model among patients receiving palliative radiotherapy (RT) that identifies those patients with short (< 3 months) and long (> 1 year) LEs.
The records of 862 patients with metastatic cancer receiving palliative RT at the Dana-Farber/Brigham and Women's Cancer Center between June 2008 and July 2011 were retrospectively reviewed. Cox proportional hazards models were used to evaluate established and potential clinical predictors of LE to construct a model predicting LE of < 3 months and > 1 year.
The median survival was 5.6 months. On multivariate analysis, factors found to be significantly associated with a shorter LE were cancer type (lung and other vs breast and prostate), older age (> 60 years vs ≤ 60 years), liver metastases, Eastern Cooperative Oncology Group performance status (2-4 vs 0-1), hospitalizations within 3 months before palliative RT (0 vs ≥ 1), and prior palliative chemotherapy courses (≥ 2 vs 0-1). Patients were divided into 3 groups with distinct median survivals: group A (those with 0-1 risk factors), 19.9 months (95% confidence interval [95% CI, 13.9 months-31.1months]); group B (those with 2-4 risk factors), 5.0 months (95% CI, 4.3 months -5.6 months); and group C (those with 5-6 risk factors), 1.7 months (95% CI, 1.2 months-2.1 months).
The TEACHH model (type of cancer, Eastern Cooperative Oncology Group performance status, age, prior palliative chemotherapy, prior hospitalizations, and hepatic metastases) divides patients receiving palliative RT into 3 distinct LE groups at clinically informative extremes of the LE spectrum. It holds promise to assist radiation oncologists in tailoring palliative therapies to a patient's LE. Cancer 2013. © 2013 American Cancer Society.
C-reactive protein (CRP) has been associated with outcomes in patients with metastatic adenocarcinoma of the prostate. Associations between prostate adenocarcinoma-specific endpoints and CRP in patients who are treated for localized disease remain unknown.
In total, 206 patients who received radiation therapy for adenocarcinoma of the prostate had at least 1 CRP measured in follow-up and were analyzed. The primary outcome was biochemical failure-free survival. In addition, associations were examined between CRP and prostate-specific antigen (PSA).
On univariate analysis, higher CRP levels were associated significantly with shorter biochemical failure-free survival for patients who received radiation therapy after undergoing radical prostatectomy. For patients who were managed with definitive radiation therapy alone, higher CRP levels also were associated significantly with shorter biochemical failure-free survival on univariate and multivariable analyses (hazard ratio, 2.03; 95% confidence interval, 1.19-3.47; P = .009). In addition, CRP levels were associated significantly with PSA after radical prostatectomy for patients who had Gleason scores ≥8 (P = .037), for high-risk patients (P = .008), and for those with pretreatment PSA levels >20 ng/mL (P = .05). In patients who received definitive radiation therapy, CRP levels also were associated with PSA both for those with pretreatment PSA levels >20 ng/mL (P < .001), and for the intermediate-risk (P = .029) and high-risk (P = .009) subgroups.
A higher CRP level was associated with shorter biochemical failure-free survival on univariate and multivariable analyses in patients who received definitive radiation therapy. CRP was also associated with PSA in exploratory subgroups. These findings warrant further exploration in a prospectively enrolled patient cohort. Cancer 2013;. © 2013 American Cancer Society.
To derive and validate a simple predictive model for survival of patients with metastatic cancer attending a palliative radiotherapy clinic.
We described previously a model predicting survival of patients referred for palliative radiotherapy using six prognostic factors: primary cancer site, site of metastases, Karnofsky performance score (KPS), and the fatigue, appetite, and shortness of breath subscales from the Edmonton Symptom Assessment Scale. Here we simplified the model to include only three factors: primary cancer site, site of metastases, and KPS. Each factor was assigned a value proportional to its prognostic weight, and the weighted scores for each patient were summed to obtain a survival prediction score (SPS). Patients were also grouped according to their number of risk factors (NRF): nonbreast cancer, metastases other than bone, and KPS < or = 60. The three- and six- variable models were evaluated for their ability to predict survival in patients referred during a different time period and of those referred to a different cancer center.
A training set of 395 patients, a temporal validation set of 445 patients, and an external validation set of 467 patients were used. The ability of the three- and six-variable models to separate patients into three prognostic groups and to predict their survival was similar using both SPS and NRF methods in the training, temporal, and external validation data sets. There was no statistically significant difference in the performance of the models.
The three-variable NRF model is preferred because of its relative simplicity.