Use of Machine Learning and Lay Care Coaches to Increase Advance Care Planning Conversations for Patients With Metastatic Cancer

PURPOSE

Patients with metastatic cancer benefit from advance care planning (ACP) conversations. We aimed to improve ACP using a computer model to select high-risk patients, with shorter predicted survival, for conversations with providers and lay care coaches. Outcomes included ACP documentation frequency and end-of-life quality measures.

METHODS

In this study of a quality improvement initiative, providers in four medical oncology clinics received Serious Illness Care Program training. Two clinics (thoracic/genitourinary) participated in an intervention, and two (cutaneous/sarcoma) served as controls. ACP conversations were documented in a centralized form in the electronic medical record. In the intervention, providers and care coaches received weekly e-mails highlighting upcoming clinic patients with < 2 year computer-predicted survival and no prior prognosis documentation. Care coaches contacted these patients for an ACP conversation (excluding prognosis). Providers were asked to discuss and document prognosis.

RESULTS

In the four clinics, 4,968 clinic visits by 1,251 patients met inclusion criteria (metastatic cancer with no prognosis previously documented). In their first visit, 28% of patients were high-risk (< 2 year predicted survival). Preintervention, 3% of both intervention and control clinic patients had ACP documentation during a visit. By intervention end (February 2021), 35% of intervention clinic patients had ACP documentation compared with 3% of control clinic patients. Providers' prognosis documentation rate also increased in intervention clinics after the intervention (2%-27% in intervention clinics, P < .0001; 0%-1% in control clinics). End-of-life care intensity was similar in intervention versus control clinics, but patients with ≥ 1 provider ACP edit met fewer high-intensity care measures (P = .04).

CONCLUSION

Combining a computer prognosis model with care coaches increased ACP documentation.

Improving lung cancer screening rates among patients with head and neck cancer in a radiation oncology clinic

Background

The United States Preventive Services Task Force (USPSTF) recommends lung cancer screening via annual low dose computed tomography (LDCT) for high risk patients. Despite the strong evidence of a mortality benefit from several randomized clinical trials, rates of lung cancer screening remain low. We plan to assess how screening guidelines are implemented in a radiation oncology clinic for patients with head and neck cancer.

Methods

A single institution, retrospective chart review was used to identify patients with head and neck cancer seen in a radiation oncology clinic who were potentially eligible for lung cancer screening under the current USPSTF guidelines. Patients who were potentially screening-eligible were enrolled in a phone survey to assess their knowledge about lung cancer screening and willingness to be screened.

Results

Of the 184 patients with head and neck cancer seen in the clinic, 8 (4%) patients were eligible for lung cancer screening under the previous USPSTF recommendations, including 1 (0.5%) patient already being screened. One patient (0.5%) became eligible under the expanded guidelines. All 184 patients had smoking history documented. Of the 87 current or former smokers, there were 24 (28%) who did not have pack-years documented; of the 82 former smokers, there were 8 (10%) who did not have quit date documented. Among the 16 phone survey participants (response rate: 70%) only 6 (38%) were aware there is a way to screen for lung cancer and 12 (75%) patients would be interested in screening if they are found to be eligible.

Conclusions

These findings highlight a potential opportunity to increase rates of lung cancer screening among patients with head and neck cancer by both enhancing provider awareness as well as patient education at the community level.

Considerations in the reliability and fairness audits of predictive models for advance care planning

Multiple reporting guidelines for artificial intelligence (AI) models in healthcare recommend that models be audited for reliability and fairness. However, there is a gap of operational guidance for performing reliability and fairness audits in practice. Following guideline recommendations, we conducted a reliability audit of two models based on model performance and calibration as well as a fairness audit based on summary statistics, subgroup performance and subgroup calibration. We assessed the Epic End-of-Life (EOL) Index model and an internally developed Stanford Hospital Medicine (HM) Advance Care Planning (ACP) model in 3 practice settings: Primary Care, Inpatient Oncology and Hospital Medicine, using clinicians' answers to the surprise question (“Would you be surprised if [patient X] passed away in [Y years]?”) as a surrogate outcome. For performance, the models had positive predictive value (PPV) at or above 0.76 in all settings. In Hospital Medicine and Inpatient Oncology, the Stanford HM ACP model had higher sensitivity (0.69, 0.89 respectively) than the EOL model (0.20, 0.27), and better calibration (O/E 1.5, 1.7) than the EOL model (O/E 2.5, 3.0). The Epic EOL model flagged fewer patients (11%, 21% respectively) than the Stanford HM ACP model (38%, 75%). There were no differences in performance and calibration by sex. Both models had lower sensitivity in Hispanic/Latino male patients with Race listed as “Other.” 10 clinicians were surveyed after a presentation summarizing the audit. 10/10 reported that summary statistics, overall performance, and subgroup performance would affect their decision to use the model to guide care; 9/10 said the same for overall and subgroup calibration. The most commonly identified barriers for routinely conducting such reliability and fairness audits were poor demographic data quality and lack of data access. This audit required 115 person-hours across 8–10 months. Our recommendations for performing reliability and fairness audits include verifying data validity, analyzing model performance on intersectional subgroups, and collecting clinician-patient linkages as necessary for label generation by clinicians. Those responsible for AI models should require such audits before model deployment and mediate between model auditors and impacted stakeholders.

Posttreatment FDG-PET/CT Hopkins criteria predict locoregional recurrence after definitive radiotherapy for oropharyngeal squamous cell carcinoma

BACKGROUND

Metabolic response assessment for oropharyngeal squamous cell carcinoma (OPSCC) aids in identifying locoregional persistence/recurrence (LRR). The Hopkins Criteria are a standardized qualitative response assessment system using posttreatment FDG-PET/CT.

METHODS

We conducted a retrospective cohort study of patients with node-positive OPSCC treated with definitive (chemo)radiotherapy. We assessed Hopkins Criteria performance for LRR, then developed and validated a competing-risks model.

RESULTS

Between 2004 and 2018, 259 patients were included with median follow-up of 43 months. The Hopkins Criteria sensitivity, specificity, negative predictive value, and accuracy were 68%, 88%, 95%, and 85%. The 36-month cumulative incidence of LRR was greater with positive scores (45% vs. 5%, HR 12.60, p < 0.001). PET/CTs performed ≤10 weeks after radiotherapy were associated with a four-fold increase in pathologically negative biopsies/surgeries (36% vs. 9%, p = 0.03). The AUC for LRR was 0.89 using a model integrating the Hopkins score.

CONCLUSIONS

The Hopkins Criteria predict LRR with high accuracy for OPSCC response assessment.

Use of systemic cancer treatments based on a validated survival prediction model in metastatic cancer

e13515

Background: Use of systemic anti-cancer treatment near the end of life (EOL) is recognized as a low value practice with limited benefit to patients. Machine learning (ML) models that identify patients in close proximity to death can help prospectively assess oncology practice of systemic therapy use. We hypothesized that systemic therapy use would be higher based on predicted survival compared with actual survival. Methods: We calculated prevalence of systemic therapy use based on predicted and actual survival among patients with metastatic cancer at Stanford Healthcare from 2008-2019. Patients were included if they were in the test set of the ML model, had an eligible outpatient oncology clinic visit for which a predicted survival was calculated and were deceased . Median predicted survival was calculated from the ML model at each outpatient oncology visit and treatment was linked to a visit date if within 14 days of each other. Prevalence of systemic therapy was calculated for patients with a predicted or actual survival of < 6 months, 6-12 months, 12-18 months and 18-24 months. The five categories of treatment were: chemotherapy, targeted/antibody, hormone, immunotherapy, and other. Results: A total of 951 deceased patients who received anticancer treatment are included and a total of 21,283 doses of treatment were administered with a mean of 22 doses per patient. The median age at metastatic cancer diagnosis was 58 years, 53% of patients were female and most patients identified as White (55%) or Asian (23%). The most common disease groups were gastrointestinal (21.6%), thoracic (18.6%) and breast (14.9%). Overall, the use of different treatment types did not differ based on either predicted or actual survival (Table). In all the survival groupings, chemotherapy remained the predominant medication type, however with a trend of decreasing use with longer predicted and actual survival. Conclusions: The use of cancer medications and the type of medication given did not change based on predicted or actual survival in a large group of patients with metastatic cancer. There was a trend of decreasing chemotherapy use with longer prognosis. Further investigation into use in time intervals closer to (predicted or actual) death and inclusion of those who did not receive any systemic therapy are underway.[Table: see text]

Accuracy of Electronic Medical Record Follow-Up Data for Estimating the Survival Time of Patients With Cancer

PURPOSE

For real-world evidence, it is convenient to use routinely collected data from the electronic medical record (EMR) to measure survival outcomes. However, patients can become lost to follow-up, causing incomplete data and biased survival time estimates. We quantified this issue for patients with metastatic cancer seen in an academic health system by comparing survival estimates from EMR data only and from EMR data combined with high-quality cancer registry data.

MATERIALS AND METHODS

Patients diagnosed with metastatic cancer from 2008 to 2014 were included in this retrospective study. Patients who were diagnosed with cancer or received their initial treatment within our system were included in the institutional cancer registry and this study. Overall survival was calculated using the Kaplan-Meier method. Survival curves were generated in two ways: using EMR follow-up data alone and using EMR data supplemented with data from the Stanford Cancer Registry/California Cancer Registry.

RESULTS

Four thousand seventy-seven patients were included. The median follow-up using EMR + Cancer Registry data was 19.9 months, and the median follow-up in surviving patients was 67.6 months. There were 1,301 deaths recorded in the EMR and 3,140 deaths recorded in the Cancer Registry. The median overall survival from the date of cancer diagnosis using EMR data was 58.7 months (95% CI, 54.2 to 63.2); using EMR + Cancer Registry data, it was 20.8 months (95% CI, 19.6 to 22.3). A similar pattern was seen using the date of first systemic therapy or date of first hospital admission as the baseline date.

CONCLUSION

Using EMR data alone, survival time was overestimated compared with EMR + Cancer Registry data.

Uncovering interpretable potential confounders in electronic medical records

Randomized clinical trials (RCT) are the gold standard for informing treatment decisions. Observational studies are often plagued by selection bias, and expert-selected covariates may insufficiently adjust for confounding. We explore how unstructured clinical text can be used to reduce selection bias and improve medical practice. We develop a framework based on natural language processing to uncover interpretable potential confounders from text. We validate our method by comparing the estimated hazard ratio (HR) with and without the confounders against established RCTs. We apply our method to four cohorts built from localized prostate and lung cancer datasets from the Stanford Cancer Institute and show that our method shifts the HR estimate towards the RCT results. The uncovered terms can also be interpreted by oncologists for clinical insights. We present this proof-of-concept study to enable more credible causal inference using observational data, uncover meaningful insights from clinical text, and inform high-stakes medical decisions.

Randomized clinical trials are often plagued by selection bias, and expert-selected covariates may insufficiently adjust for confounding factors. Here, the authors develop a framework based on natural language processing to uncover interpretable potential confounders from text.

Acute and Late Esophageal Toxicity Following Stereotactic Ablative Radiotherapy to Thoracic Tumors near or Abutting the Esophagus

PURPOSE

To evaluate the incidence of acute and late esophageal toxicity in patients with thoracic tumors near or abutting the esophagus treated with stereotactic ablative radiotherapy (SABR).

METHODS AND MATERIALS

Among patients with thoracic tumors treated with SABR, we identified those with tumors near or abutting the esophagus. Using the linear-quadratic model with an α/ß ratio of 10, we determined the correlation between dosimetric parameters and esophageal toxicity graded using the Common Terminology Criteria for Adverse Events (CTCAE), version 5.0.

RESULTS

Out of 2200 patients treated with thoracic SABR, 767 patients were analyzable for esophageal dosimetry. We identified 55 patients with tumors near the esophagus (52 evaluable for esophagitis grade), 28 with PTV overlapping the esophagus. Median follow-up and overall survival were 16 and 23 months respectively. Thirteen patients (25%) developed temporary grade 2 acute esophageal toxicity, 11 (85%) of whom had PTV overlapping the esophagus. Symptoms resolved within 1-3 months in 12 patients, and 6 months in all patients. No grade 3-5 toxicity was observed. Only 3 patients (6%) developed late or persistent grade 2 dysphagia or dyspepsia of uncertain relationship to SABR. Cumulative incidence of acute esophagitis was 15% and 25% at 14 days and 60 days respectively. Acute toxicity correlated on univariate analysis with esophageal D_max, D_1cc, D_2cc, D_max/D_prescription and whether the PTV was overlapping the esophagus. Esophageal D_max (BED₁₀) < 62 Gy, D_1cc (BED₁₀) < 48 Gy, D_2cc (BED₁₀) < 43 Gy, and D_max/D_prescription < 85% was associated with <20% risk of grade 2 acute esophagitis. Only 2 local recurrences occurred.

CONCLUSIONS

Although 25% of patients with tumors near the esophagus developed acute esophagitis (39% of those with PTV overlapping the esophagus), these toxicities were all grade 2 and all temporary. This suggests the safety and efficacy of thoracic SABR for tumors near or abutting the esophagus when treating with high conformity and sharp dose gradients.

Use of a computer model and care coaches to increase advance care planning conversations for patients with metastatic cancer

8

Background: Patients with metastatic cancer benefit from advance care planning (ACP) conversations. Despite initiatives which train providers to have ACP conversations using the serious illness care program (SICP) conversation guide, few patients have a documented prognosis discussion due to busy clinic schedules and difficulty in deciding the right times to have such conversations. We designed an intervention to improve ACP by incorporating a validated computer model to identify patients at high risk for mortality in combination with lay care coaches. We investigated whether this would improve end of life quality measures. Methods: Four Stanford clinics were included in this pilot; all received SICP training. Two clinics (thoracic and genitourinary) underwent the intervention (computer model + care coach), and two clinics (sarcoma and cutaneous) served as the control. For providers in the intervention, an email was sent every Sunday listing the metastatic cancer patients who would be seen in clinic the following week and a predicted prognosis generated by the model. A lay care coach contacted patients with a predicted survival ≤2 years to have an ACP conversation with them. After, the care coach notified the provider to suggest discussion regarding prognosis with the patient. Criteria for a patient visit to be included in the analysis were: age ≥18, established patient, has sufficient EMR data for computer model, and no prior prognosis documentation. The primary outcome was documentation of prognosis in the ACP form by the end of the week following the clinic visit. Results: 5330 visits in 1298 unique patients met the inclusion criteria. Median age was 67 (range 19-97); 790 male, 508 female. 1970 visits were with patients with ≤2 year predicted survival. Prognosis discussion was documented by providers in the ACP form for 8.1% of intervention visits compared to 0.07% of control visits (p=0.001 in mixed effects model). Of the 1298 unique patients, 84 were deceased by December 2020. 41.7% died in the hospital. 59.5% were enrolled in hospice prior to death, and 19.0% were hospitalized in the ICU ≤14 days prior to death. Of deceased patients with ACP form prognosis documentation, 5.0% had ≥2 hospitalizations in the 30 days before death compared to 23.4% of deceased patients with no prognosis documented (p=0.10). For ≥ 2 ER visits in the 30 days before death, the proportions were 5.0% and 20.3% (p=0.17). Conclusions: This pilot study supports that our intervention is associated with higher rates of prognosis discussions and documentation. There was a trend towards better quality of end of life care as noted by higher rates of hospice enrollment and less intensive care at end of life. These results merit further investigation as a means to improve goal-concordant care and ensure appropriate care for cancer patients at the end of life.

Natural Language Processing to Identify Cancer Treatments With Electronic Medical Records

PURPOSE

Knowing the treatments administered to patients with cancer is important for treatment planning and correlating treatment patterns with outcomes for personalized medicine study. However, existing methods to identify treatments are often lacking. We develop a natural language processing approach with structured electronic medical records and unstructured clinical notes to identify the initial treatment administered to patients with cancer.

METHODS

We used a total number of 4,412 patients with 483,782 clinical notes from the Stanford Cancer Institute Research Database containing patients with nonmetastatic prostate, oropharynx, and esophagus cancer. We trained treatment identification models for each cancer type separately and compared performance of using only structured, only unstructured (bag-of-words, doc2vec, fasttext), and combinations of both (structured + bow, structured + doc2vec, structured + fasttext). We optimized the identification model among five machine learning methods (logistic regression, multilayer perceptrons, random forest, support vector machines, and stochastic gradient boosting). The treatment information recorded in the cancer registry is the gold standard and compares our methods to an identification baseline with billing codes.

RESULTS

For prostate cancer, we achieved an f1-score of 0.99 (95% CI, 0.97 to 1.00) for radiation and 1.00 (95% CI, 0.99 to 1.00) for surgery using structured + doc2vec. For oropharynx cancer, we achieved an f1-score of 0.78 (95% CI, 0.58 to 0.93) for chemoradiation and 0.83 (95% CI, 0.69 to 0.95) for surgery using doc2vec. For esophagus cancer, we achieved an f1-score of 1.0 (95% CI, 1.0 to 1.0) for both chemoradiation and surgery using all combinations of structured and unstructured data. We found that employing the free-text clinical notes outperforms using the billing codes or only structured data for all three cancer types.

CONCLUSION

Our results show that treatment identification using free-text clinical notes greatly improves upon the performance using billing codes and simple structured data. The approach can be used for treatment cohort identification and adapted for longitudinal cancer treatment identification.

Automated model versus treating physician for predicting survival time of patients with metastatic cancer

OBJECTIVE

Being able to predict a patient's life expectancy can help doctors and patients prioritize treatments and supportive care. For predicting life expectancy, physicians have been shown to outperform traditional models that use only a few predictor variables. It is possible that a machine learning model that uses many predictor variables and diverse data sources from the electronic medical record can improve on physicians' performance. For patients with metastatic cancer, we compared accuracy of life expectancy predictions by the treating physician, a machine learning model, and a traditional model.

MATERIALS AND METHODS

A machine learning model was trained using 14 600 metastatic cancer patients' data to predict each patient's distribution of survival time. Data sources included note text, laboratory values, and vital signs. From 2015-2016, 899 patients receiving radiotherapy for metastatic cancer were enrolled in a study in which their radiation oncologist estimated life expectancy. Survival predictions were also made by the machine learning model and a traditional model using only performance status. Performance was assessed with area under the curve for 1-year survival and calibration plots.

RESULTS

The radiotherapy study included 1190 treatment courses in 899 patients. A total of 879 treatment courses in 685 patients were included in this analysis. Median overall survival was 11.7 months. Physicians, machine learning model, and traditional model had area under the curve for 1-year survival of 0.72 (95% CI 0.63-0.81), 0.77 (0.73-0.81), and 0.68 (0.65-0.71), respectively.

CONCLUSIONS

The machine learning model's predictions were more accurate than those of the treating physician or a traditional model.

Postoperative Observation Versus Radiotherapy for Pathologic N1 Oral Cavity Squamous Cell Carcinoma

OBJECTIVES

To investigate the benefit of postoperative radiotherapy (PORT) for low-volume (pN1) nodal disease after resection of oral cavity squamous cell carcinoma.

MATERIALS AND METHODS

The National Cancer Database was queried for adults with nonmetastatic squamous cell carcinoma of the oral cavity treated by surgical resection with pathologic stage T1-2 N0-2 (American Joint Committee on Cancer 7th edition) and with the maximal exclusion of standard indications for PORT. Overall survival was compared within pN1 for observation versus PORT and then compared for pN1 versus pN0 and versus pN2 stratified by receipt of observation or PORT. Multivariable Cox regression was used to adjust for potential confounders between PORT and survival, including comorbidity and age.

RESULTS

Overall 5017 pN0, 530 pN1, and 253 pN2 patients were identified, of whom 9%, 35%, and 64% received PORT, respectively. Within the pN1 cohort, PORT was associated with improved survival versus observation (adjusted hazard ratio, 0.66; 95% confidence interval, 0.46-0.97; P=0.03). Among observed patients, the prognosis of pN1 was equivalent to pN2 and inferior to pN0; in contrast, among patients treated with PORT, the prognosis of pN1 was equivalent to pN0 and superior to pN2. Without PORT, pN1 remained an adverse risk factor relative to pN0 regardless of the depth of invasion, lymph node size, lymph node location, and extent of lymph node dissection.

CONCLUSIONS

PORT was associated with a survival benefit compared with observation. Notably, pN1 was an adverse risk factor relative to pN0 if, and only if, patients did not receive PORT, suggesting pN1 by itself may be an indication for PORT.

Prolongation of definitive head and neck cancer radiotherapy: Survival impact and predisposing factors

BACKGROUND AND PURPOSE

To quantify the survival impact of prolongation of definitive radiotherapy (RT) for head and neck cancer in a national, modern cohort, and to identify predictive factors for prolongation.

MATERIALS AND METHODS

The National Cancer Database was queried for adults with non-metastatic cancer of the nasopharynx, oropharynx, larynx, or hypopharynx diagnosed 2004-2015, treated with definitive RT to 66-70 Gy in 30-35 fractions at 2-2.2 Gy per fraction. Multivariable Cox regression and propensity score matching were used to model the survival impact of RT prolongation, adjusting for potential confounders such as age and comorbidity. Predictors of RT prolongation were identified using multivariable multinomial logistic regression.

RESULTS

In total, 36,367 patients were identified. As a continuous variable, RT prolongation increased the relative hazard of death by 2% per day (P < .0001). In the matched cohorts, patients with short (4-8 days) or long prolongation (>8 days) had lower absolute 4-year overall survival by 4% and 12%, respectively (P < .0001), while prolongation of 1-3 days was not significantly adverse. Major predictors of increased risk of prolongation were administration of systemic therapy, baseline comorbidity, lack of private insurance, and tumor/nodal stage. Conversely, higher facility volume was significantly protective, with a 55% lower risk of long prolongation within the topmost quartile (>11.5 patients/year).

CONCLUSION

RT prolongation, especially >8 days, is significantly deleterious. Systemic therapy and facility volume were major predictors. Early identification of patients at increased risk of treatment interruptions may facilitate implementation of preventive measures.

Early response evaluation using primary tumor and nodal imaging features to predict progression-free survival of locally advanced non-small cell lung cancer

Prognostic biomarkers that can reliably predict early disease progression of non-small cell lung cancer (NSCLC) are needed for identifying those patients at high risk for progression, who may benefit from more intensive treatment. In this work, we aimed to identify an imaging signature for predicting progression-free survival (PFS) of locally advanced NSCLC. Methods: This retrospective study included 82 patients with stage III NSCLC treated with definitive chemoradiotherapy for whom both baseline and mid-treatment PET/CT scans were performed. They were randomly placed into two groups: training cohort (n=41) and testing cohort (n=41). All primary tumors and involved lymph nodes were delineated. Forty-five quantitative imaging features were extracted to characterize the tumors and involved nodes at baseline and mid-treatment as well as differences between two scans performed at these two points. An imaging signature was developed to predict PFS by fitting an L1-regularized Cox regression model. Results: The final imaging signature consisted of three imaging features: the baseline tumor volume, the baseline maximum distance between involved nodes, and the change in maximum distance between the primary tumor and involved nodes measured at two time points. According to multivariate analysis, the imaging model was an independent prognostic factor for PFS in both the training (hazard ratio [HR], 1.14, 95% confidence interval [CI], 1.04-1.24; P = 0.003), and testing (HR, 1.21, 95% CI, 1.10-1.33; P = 0.048) cohorts. The imaging signature stratified patients into low- and high-risk groups, with 2-year PFS rates of 61.9% and 33.2%, respectively (P = 0.004 [log-rank test]; HR, 4.13, 95% CI, 1.42-11.70) in the training cohort, as well as 43.8% and 22.6%, respectively (P = 0.006 [log-rank test]; HR, 3.45, 95% CI, 1.35-8.83) in the testing cohort. In both cohorts, the imaging signature significantly outperformed conventional imaging metrics, including tumor volume and SUV_max value (C-indices: 0.77-0.79 for imaging signature, and 0.53-0.73 for conventional metrics). Conclusions: Evaluation of early treatment response by combining primary tumor and nodal imaging characteristics may improve the prediction of PFS of locally advanced NSCLC patients.

Predicting per-lesion local recurrence in locally advanced non-small cell lung cancer following definitive radiation therapy using pre- and mid-treatment metabolic tumor volume

BACKGROUND

We evaluated whether pre- and mid-treatment metabolic tumor volume (MTV) predicts per lesion local recurrence (LR) in patients treated with definitive radiation therapy (RT, dose≥60 Gy) for locally advanced non-small cell lung cancer (NSCLC).

METHODS

We retrospectively reviewed records of patients with stage III NSCLC treated from 2006 to 2018 with pre- and mid-RT PET-CT. We measured the MTV of treated lesions on the pre-RT (MTV_pre) and mid-RT (MTV_mid) PET-CT. LR was defined per lesion as recurrence within the planning target volume. Receiver operating characteristic (ROC) curves, cumulative incidence rates, and uni- and multivariable (MVA) competing risk regressions were used to evaluate the association between MTV and LR.

RESULTS

We identified 111 patients with 387 lesions (112 lung tumors and 275 lymph nodes). Median age was 68 years, 69.4% were male, 46.8% had adenocarcinoma, 39.6% had squamous cell carcinoma, and 95.5% received concurrent chemotherapy. Median follow-up was 38.7 months. 3-year overall survival was 42.3%. 3-year cumulative incidence of LR was 26.8% per patient and 11.9% per lesion. Both MTV_pre and MTV_mid were predictive of LR by ROC (AUC = 0.71 and 0.76, respectively) and were significantly associated with LR on MVA (P = 0.004 and P = 7.1e-5, respectively). Among lesions at lower risk of LR based on MTV_pre, higher MTV_mid was associated with LR (P = 0.001).

CONCLUSION

Per-lesion, larger MTV_pre and MTV_mid predicted for increased risk of LR. MTV_mid was more highly predictive of LR than MTV_pre and if validated may allow for further discrimination of high-risk lesions at mid-RT informing dose painting strategies.

Tumor Subregion Evolution-Based Imaging Features to Assess Early Response and Predict Prognosis in Oropharyngeal Cancer

The incidence of oropharyngeal squamous cell carcinoma (OPSCC) has been rapidly increasing. Disease stage and smoking history are often used in current clinical trials to select patients for deintensification therapy, but these features lack sufficient accuracy for predicting disease relapse. Our purpose was to develop an imaging signature to assess early response and predict outcomes of OPSCC. Methods: We retrospectively analyzed 162 OPSCC patients treated with concurrent chemoradiotherapy, equally divided into separate training and validation cohorts with similar clinical characteristics. A robust consensus clustering approach was used to spatially partition the primary tumor and involved lymph nodes into subregions (i.e., habitats) based on 18F-FDG PET and contrast CT imaging. We proposed quantitative image features to characterize the temporal volumetric change of the habitats and peritumoral/nodal tissue between baseline and midtreatment. The reproducibility of these features was evaluated. We developed an imaging signature to predict progression-free survival (PFS) by fitting an L1-regularized Cox regression model. Results: We identified 3 phenotypically distinct intratumoral habitats: metabolically active and heterogeneous, enhancing and heterogeneous, and metabolically inactive and homogeneous. The final Cox model consisted of 4 habitat evolution-based features. In both cohorts, this imaging signature significantly outperformed traditional imaging metrics, including midtreatment metabolic tumor volume for predicting PFS, with a C-index of 0.72 versus 0.67 (training) and 0.66 versus 0.56 (validation). The imaging signature stratified patients into high-risk versus low-risk groups with 2-y PFS rates of 59.1% versus 89.4% (hazard ratio, 4.4; 95% confidence interval, 1.4-13.4 [training]) and 61.4% versus 87.8% (hazard ratio, 4.6; 95% confidence interval, 1.7-12.1 [validation]). The imaging signature remained an independent predictor of PFS in multivariable analysis adjusting for stage, human papillomavirus status, and smoking history. Conclusion: The proposed imaging signature allows more accurate prediction of disease progression and, if prospectively validated, may refine OPSCC patient selection for risk-adaptive therapy.

A Radiation Oncology-Specific Automated Trigger Indicator Tool for High-Risk, Near-Miss Safety Events

PURPOSE

Error detection in radiation oncology relies heavily on voluntary reporting, and many adverse events and near misses likely go undetected. Trigger tools use existing data in patient charts to identify otherwise-unaccounted-for events and have been successfully employed in other areas of medicine. We developed an automated radiation oncology-specific trigger tool and validated it against near-miss data from a high-volume incident learning system (ILS).

METHODS AND MATERIALS

Twenty triggers were derived from an electronic radiation oncology information system. Data from the systems over an approximately 3.5-year period were split randomly into training and test sets. The probability of a high-grade (grade 3-4) near miss for each treatment course in the training set was estimated using a regularized logistic regression model. The predictive model was applied to the test set. Records for 25 flagged treatment courses with an ILS entry were reviewed to explore the association between triggers and near misses, and 25 flagged courses without an ILS entry were reviewed to detect unreported near misses.

RESULTS

Of the 3159 treatment courses analyzed, 357 had a grade 3 to 4 ILS entry; 2210 courses composed the training set, and the test set had 949 courses. Areas under the curve on the training and test sets were 0.650 and 0.652, respectively. Of 20 triggers, 9 reached statistical significance on univariate analysis. Fifty percent of the 25 treatment courses in the test set with the highest predicted likelihood of a high-grade near miss with an ILS entry had a direct relationship between the triggers and the near miss. Review of the 25 treatment courses with the highest predicted likelihood of high-grade near miss without an ILS entry found 2 unreported near-miss events.

CONCLUSIONS

The radiation oncology-specific automated trigger tool performed modestly and identified additional treatment courses with near-miss events. Radiation oncology trigger tools deserve further exploration.

Predictors of Respiratory Decline Following Stereotactic Ablative Radiotherapy to Multiple Lung Tumors

INTRODUCTION

Stereotactic ablative radiotherapy (SABR) is highly effective at controlling early stage primary lung cancer and lung metastases. Although previous studies have suggested that treating multiple lung tumors with SABR is safe, post-treatment changes in respiratory function have not been analyzed in detail.

PATIENTS AND METHODS

We retrospectively identified patients with 2 or more primary lung cancers or lung metastases treated with SABR and analyzed clinical outcomes and predictors of toxicity. We defined a composite respiratory decline endpoint to include increased oxygen requirement, increased dyspnea scale, or death from respiratory failure not owing to disease progression.

RESULTS

A total of 86 patients treated with SABR to 203 lung tumors were analyzed. A total of 21.8% and 41.8% of patients developed composite respiratory decline at 2 and 4 years, respectively. When accounting for intrathoracic disease progression, 12.7% of patients developed composite respiratory decline at 2 years. Of the patients, 7.9% experienced grade 2 or greater radiation pneumonitis. No patient- or treatment-related factor predicted development of respiratory decline. The median overall survival was 46.9 months, and the median progression-free survival was 14.8 months. The cumulative incidence of local failure was 9.7% at 2 years.

CONCLUSION

Although our results confirm that SABR is an effective treatment modality for patients with multiple lung tumors, we observed a high rate of respiratory decline after treatment, which may be owing to a combination of treatment and disease effects. Future studies may help to determine ways to avoid pulmonary toxicity from SABR.

Automated Survival Prediction in Metastatic Cancer Patients Using High-Dimensional Electronic Medical Record Data

BACKGROUND

Oncologists use patients' life expectancy to guide decisions and may benefit from a tool that accurately predicts prognosis. Existing prognostic models generally use only a few predictor variables. We used an electronic medical record dataset to train a prognostic model for patients with metastatic cancer.

METHODS

The model was trained and tested using 12 588 patients treated for metastatic cancer in the Stanford Health Care system from 2008 to 2017. Data sources included provider note text, labs, vital signs, procedures, medication orders, and diagnosis codes. Patients were divided randomly into a training set used to fit the model coefficients and a test set used to evaluate model performance (80%/20% split). A regularized Cox model with 4126 predictor variables was used. A landmarking approach was used due to the multiple observations per patient, with t0 set to the time of metastatic cancer diagnosis. Performance was also evaluated using 399 palliative radiation courses in test set patients.

RESULTS

The C-index for overall survival was 0.786 in the test set (averaged across landmark times). For palliative radiation courses, the C-index was 0.745 (95% confidence interval [CI] = 0.715 to 0.775) compared with 0.635 (95% CI = 0.601 to 0.669) for a published model using performance status, primary tumor site, and treated site (two-sided P < .001). Our model's predictions were well-calibrated.

CONCLUSIONS

The model showed high predictive performance, which will need to be validated using external data. Because it is fully automated, the model can be used to examine providers' practice patterns and could be deployed in a decision support tool to help improve quality of care.