Article Tools
ORIGINAL CONTRIBUTIONS
Care Delivery
Article Tools
Identification of Patient-Reported Outcome Phenotypes Among Oncology Patients With Palliative Care Needs
, MD, MSCE1,2,3; Kelly D. Getz , PhD, MPH1,4,5; Jesse Y. Hsu , PhD4; Antonia V. Bennett, PhD6; Samuel U. Takvorian , MD, MSHP1,2; Arif H. Kamal , MD, MBA, MHS7; and Angela DeMichele , MD, MSCE1 Show More
Abstract
Despite evidence-based guidelines recommending early palliative care, it remains unclear how to identify and refer oncology patients, particularly in settings with constrained access to palliative care. We hypothesize that patient-reported outcome (PRO) data can be used to characterize patients with palliative care needs. To determine if PRO data can identify latent phenotypes that characterize indications for specialty palliative care referral.
We conducted a retrospective study of self-reported symptoms on the Edmonton Symptom Assessment System collected from solid tumor oncology patients (n = 745) referred to outpatient palliative care. Data were collected as part of routine clinical care from October 2012 to March 2018 at eight community and academic sites. We applied latent profile analysis to identify PRO phenotypes and examined the association of phenotypes with clinical and demographic characteristics using multinomial logistic regression.
We identified four PRO phenotypes: (1) Low Symptoms (n = 295, 39.6%), (2) Moderate Pain/Fatigue + Mood (n = 180, 24.2%), (3) Moderate Pain/Fatigue + Appetite + Dyspnea (n = 201, 27.0%), and (4) High Symptoms (n = 69, 9.3%). In a secondary analysis of 421 patients, we found that two brief items assessing social and existential needs aligned with higher severity symptom and psychological distress phenotypes.
Oncology patients referred to outpatient palliative care in a real-world setting can be differentiated into clinically meaningful phenotypes using brief, routinely collected PRO measures. Latent modeling provides a mechanism to use patient-reported data on a population level to identify distinct subgroups of patients with unmet palliative needs.
Early outpatient palliative care for patients with advanced solid tumors improves healthcare outcomes.1-6 Several randomized trials have consistently shown significant improvements in quality of life, symptom burden, end-of-life care, and a potential survival benefit with the addition of early specialty palliative care to standard oncology treatment for patients with advanced cancer.1-5,7 However, palliative care is not well-integrated into oncology practice despite clear recommendations for timely referral.8,9 One major barrier to referral is the lack of clearly defined referral criteria for specialty palliative care, particularly in the setting of constrained access.10-13 Current approaches for identifying patients with outpatient palliative care needs rely on disease characteristics, patient symptoms, clinician intuition, or measures of a patient's effect on health system costs, such as unplanned emergency room visits or hospital readmission.14-17 Referral criteria would optimally consider direct patient report measured through standardized patient-reported outcomes (PROs), which may improve the identification of symptoms and timing of referral among oncology patients.
Symptom monitoring using PRO measures is uniquely suited to capture active, time-varying, and complex palliative needs directly from the patient. With the growing use of PRO symptom monitoring in oncology,18-20 there is now an opportunity to study whether self-reported symptom data can direct the timing and appropriateness of palliative care referral. However, to use PRO data on a population level to identify patients' needs and direct care, we first need to develop computational methods to interpret large PRO data sets. Latent profile analysis (LPA) is a well-validated statistical approach that lends itself to interpretation of symptom data.21-23
LPA uses responses to survey questions to identify latent subgroups, with the hypothesis that patient membership in a subgroup may predict similar risk for future outcomes or differentiate response to interventions.24 Although previous studies have examined symptom data to identify latent subgroups among oncology patients, these studies were not designed to identify subgroups of patients with actionable palliative care needs.23,25,26 The application of these studies to inform development of PROs for palliative care referral is limited because of the use of heterogeneous patient populations and convenience sampling of PRO data. Thus, the primary aim of this study is to determine if PRO data can be used to identify latent phenotypes among oncology patients that characterize indications for palliative care referral. We then determine the association of latent phenotypes with demographic and clinical characteristics.
We conducted a retrospective cross-sectional study of PROs reported by oncology patients at the time of their initial palliative care visit. Data were collected as part of routine clinical care from October 2012 to March 2018 at eight community and academic sites within the Global Palliative Care Quality Alliance (GPCQA).27 To ensure high fidelity of data collection, we excluded sites with (1) 20% or greater missingness in any one of the nine PRO items included in the primary analysis and (2) low site volume, defined as < 0.5% of the total study population. The study population (N = 894) included all adult patients (≥ 18 years old) with a solid tumor diagnosis seen at an initial palliative care visit at one of the sites within the GPCQA, and 745 patients with complete PRO data were included in the primary analysis (Appendix Fig A1, online only, CONSORT diagram).
Data were collected using the Quality Data Collection Tool (QDACT), a secure Web-based interface used by clinicians to input data from patients at the point of care.28 QDACT facilitates the structured assessment of palliative care domains and includes instruments that have been well-validated in oncology research and clinical practice. We selected the QDACT data set for several reasons: (1) a registry population representative of natural patterns of palliative care referral in the real-world setting from community and academic practices; (2) inclusion of PRO measures that assess multiple domains of palliative care, including physical, psychosocial, and existential distress; and (3) inclusion of disease-specific (tumor type and stage) and clinical (performance status and provider prognostic estimation) data elements. This study was approved by the Duke Institutional Review Board (Pro00035703).
PRO measures in the QDACT registry were selected through literature review and national quality guidelines for symptom assessment in palliative care.28,29 Demographics and clinical data are collected at the beginning of the palliative care visit and include the following categories.
Sociodemographic variables are age at time of consultation, sex, race, and partner status.
Clinical variables are tumor type, stage, referring specialty, palliative performance scale, and provider prognostic estimation.
Physical and psychological symptoms are assessed using the Edmonton Symptom Assessment System (ESAS), a widely used self-reporting tool of symptom intensity developed for the oncology patient population.30 Symptoms are rated on a 0-10 numerical rating scale, in which 0 means that the symptom is not present and 10 means that the symptom is the “worst possible” and include pain, shortness of breath, constipation, fatigue, nausea, drowsiness, appetite, depression, anxiety, and well-being.
Existential distress (“Are you at peace?”) and social distress (“At times I worry I will be a burden to my family.”) are each assessed with a single question that has been modified from existing instruments for feasibility of collection in routine clinical care.28 These questions were developed by the QDACT investigators in collaboration with the GPCQA and by consultation with national experts and then refined through multiple iterations for clinical utility based on patient and provider feedback (Appendix Table A1, online only).
We hypothesized that particular combinations of responses to the ESAS symptom questions (indicator variables) would define clinically relevant PRO phenotypes (latent subgroups) that characterize indications for specialty palliative care referral. LPA uses finite mixture models to identify unobserved latent subgroups that are defined by a set of indicator variables.24 Statistical analysis was conducted using Stata Version 15.1.
The ESAS “well-being” item was excluded because of potential measurement error that resulted from a reverse rating, that is, 0 is worst well-being and 10 is best well-being. We treated the answer option “patient unable to respond” as missing because it showed inconsistent patterns of response across symptoms for individual patients. We compared the ESAS complete case population (n = 745) with the population with missing ESAS data and found potential differences in performance status and provider-estimated prognosis (interpretation limited by missingness), but otherwise no differences in demographic and clinical characteristics (Appendix Table A2). We then proceeded with a complete case analysis.
To determine the optimal number of latent subgroups, we used an exploratory approach and compared model fit indices for increasing numbers of classes until the Akaike information criterion and Bayesian information criterion plateaued, resulting in an analysis of 2-6 classes (Appendix Fig A2). To determine the final model, lower values of Akaike information criterion and Bayesian information criterion were considered in conjunction with clinical interpretability of the subgroups to reach an optimal balance between model fit and clinical relevance. We characterized the severity and types of ESAS symptoms to describe PRO phenotypes and then assigned latent subgroup membership to each patient in the data set based on the highest predicted probability of subgroup membership (posterior probability).
Patients’ clinical and demographics characteristics were summarized as mean (standard deviation) for continuous variables and frequency (%) for categorical variables. Analysis of variance and chi-square test were used to compare continuous and categorical variables, respectively, across the PRO phenotypes. To determine construct validity of the PRO phenotypes, we evaluated whether demographics and clinical characteristics follow expected patterns across phenotypes. We hypothesized that selected characteristics (age, sex, race, relationship status, Palliative Performance Scale [PPS], provider prognostic estimation, cancer type, and stage) would associate with symptom phenotypes. We estimated these associations in a multiple multinomial logistic regression model, with the PRO phenotypes as the outcome and phenotype “Low Symptoms” as the reference group.
To explore how social and existential distress measures may inform phenotypes, we performed a secondary LPA that included two additional questions regarding these perceived states, respectively (ie, “worry about family burden” and being “not at peace”). Four hundred twenty-one patients with complete data for the additional measures contributed to this analysis. Likert scale responses were converted into binary variables (present or not present) to increase power (Appendix Table A1).
Given the high prevalence of lung cancer in the cohort, we performed a sensitivity analysis to assess whether phenotypes identified in the primary analysis were driven by the largest disease group. We excluded patients with lung cancer and examined whether the resulting phenotypes were similar to those of the full sample.
Table 1 shows clinical and demographic characteristics of the total study population (N = 745) and latent symptom phenotypes. Notably, 38.3% of patients have a PPS score of 40-60 (approximately Eastern Cooperative Oncology Group [ECOG] 2-3) and 46.3% have an estimated prognosis of < 6 months.
|
The LPA model includes nine symptom indicators: pain, shortness of breath, constipation, fatigue, nausea, drowsiness, appetite, depression, and anxiety. We identified four PRO phenotypes (Fig 1): (1) Low Symptoms (“Low”, n = 295, 39.6%), (2) Moderate Pain/Fatigue + Mood (“Mood”, n = 180, 24.2%), (3) Moderate Pain/Fatigue + Appetite + Dyspnea (“Moderate Physical”, n = 201, 27.0%), and (4) High Symptoms (“High”, n = 69, 9.3%). The four-class model achieves an optimal balance between model fit and parsimony of phenotypes for clinical application as a screening tool. Descriptively, patients in the “Low” and “High” phenotypes have nonoverlapping severity profiles. The two “Moderate” phenotypes are most clinically differentiated from each other by mood severity, in which patients in the “Mood” phenotype have higher reports of anxiety and depression.
FIG 1. (A) Latent profile model: four-class solution. Depiction of the results of the latent profile model, which shows mean ESAS severity ratings (0-10) predicted by the model for each PRO phenotype. Symptom severity is represented by a graded color scale from green (low) to red (high). (B) Mean ESAS item severity ratings by PRO phenotype. Depiction of mean ESAS severity ratings with confidence intervals observed for patients within each PRO phenotype. Patients are assigned to a PRO phenotype based on the predicted probability of class membership from the latent profile model. Patients have different profiles of symptom type and severity based on PRO phenotype membership. ESAS, Edmonton Symptom Assessment System; PRO, patient-reported outcome.
We examined associations between PRO phenotypes and patient demographic and clinical characteristics using the “Low” phenotype as the reference group (Table 2). We did not find significant differences in race, relationship status, or sex among patients by PRO phenotypes. However, conclusions regarding race and relationship status are limited by the small percentage of non-White patients in the study population and missing data in these variables. Point estimates of association with PRO phenotype and cancer type show a trend toward differences across symptom phenotypes but did not reach statistical significance. Age is significantly associated with phenotype membership. Patients of younger age (age, 25-55) are more likely to be associated with all symptomatic phenotypes and especially associate with the “Mood” phenotype compared with older patients (age, 25-55, odds ratio [OR], 5.16, 95% CI, [2.56 to 10.38]; age > 55-74; OR, 2.58, 95% CI [1.43 to 4.66]). Regarding disease stage, patients with nonmetastatic disease are more likely to be in the “High” versus “Low” phenotype at the time of palliative evaluation compared with patients with metastatic disease (“High” OR, 2.82, 95% CI [1.27 to 6.26]).
|
Point estimates of PPS scores show an increasing graded response relationship with worsening severity of PRO phenotypes (“Mood” OR, 1.99; 95% CI [1.27 to 3.10]; “Moderate Physical” OR, 2.58; 95% CI [1.69 to 3.94]; and “High” OR; 3.76, 95% CI [2.05 to 6.90]). For example, patients with a PPS 40-60 (ie, ECOG 2-3) are approximately 3.8 times more likely to be associated with the “High” phenotype compared with patients with a PPS 70-100 (ie, ECOG 0-1).
In a secondary analysis using social and existential distress measures (n = 421), we identified a four-class solution with descriptively similar phenotypes (Fig 2). Social and existential distress needs group with phenotypes that have higher severity symptoms and psychological distress. The conditional probability of being “not at peace” is .86 in the “Mood” phenotype and .73 in the “High” phenotype (compared with .33 in “Low”). The conditional probability of “worry about family burden” is .88 in the “High” phenotype and .69 in the “Mood” phenotype (compared with .38 in “Low”).
FIG 2. Latent profile model with existential and social distress measures. Depiction of symptom profiles for the four PRO phenotypes with two additional measures for being “not at peace” or having “worry about burden on family.” Symptom severity is represented by a graded color scale from green (low) to red (high). For the ESAS items, a patient's severity rating (0-10) is predicted by the latent profile model based on membership in each PRO phenotype. For the “peace” and “burden” measures, the model reports the conditional probability of a patient screening positive for distress given membership in each PRO phenotype. ESAS, Edmonton Symptom Assessment System; PRO, patient-reported outcome.
In a sensitivity analysis to assess the effect of lung cancer prevalence on the formation of symptom phenotypes, removal of patients with lung cancer (n = 211, 28.3%) did not change the latent profile model or predicted probabilities of class membership (Fig A3).
We identified four latent PRO phenotypes among solid tumor oncology patients who were referred in real-world settings for palliative care evaluation (Fig 1). The PRO phenotypes have good construct validity, with factors known to be associated with higher patient distress such as younger age and worse functional status associated with more severe symptom phenotypes.25,31,32 In a secondary analysis using two social and existential distress items, these domains grouped with symptom severity and psychological distress phenotypes. The identified PRO phenotypes are clinically recognizable (Fig 2) and potentially modifiable with clinical interventions and characterize appropriate and common indications for palliative care referral across physical, psychosocial, and existential domains. Collectively, these findings suggest that patient-reported symptoms could be used on a population level to identify patients with unmet palliative needs.
Current efforts to triage patients to specialty palliative care rely on demographic, disease, and prognostic factors. Our study examined the association of PRO phenotypes with these factors. Younger age is significantly associated with all symptomatic phenotypes and has the strongest association with the “Mood” phenotype. This finding is consistent with previous subgroup analyses of oncology patients, which found correlation of symptom occurrence with psychological distress and higher psychological distress among younger patients.23,25 Importantly, current referral criteria consider those of older age, not younger age, as an indicator for palliative care. We also examined disease severity. Patients with nonmetastatic disease are more likely to be associated with the “High” phenotype compared with “Low,” which we hypothesize reflects referral patterns. Patients with nonmetastatic disease may be referred for high symptom burden, whereas those with metastatic disease may be referred for other palliative needs such as advanced care planning or medical decision making that are not captured in this data set. Finally, there is a graded increase in association with performance status and higher severity symptoms. Taken together, these associations are clinically representative and overall provide good construct validity for the latent profile model.
In a secondary analysis using two additional questions to assess for social and existential distress, we identified a four-class model that is descriptively similar to the primary analysis with similar predicted proportions for each PRO phenotype (Fig 2). Patients are more likely to report “not being at peace” (existential distress) or “worry about being a burden to family” (social distress) in both the mood-predominant and high severity phenotypes. Furthermore, patients are most likely to report not being at peace in the “Mood” phenotype and most likely to report worry about burden on family in the “High” symptom phenotype. These groupings clinically make sense, as patients with high symptom burden likely require more caregiver support and patients with mood-predominant symptoms likely have more difficulty with coping. We hypothesize that the addition of existential and social domains to PRO-based screening for palliative needs may help tailor supportive services, such as targeting social work services to patients who report “worry about family burden.” Further research is needed to prospectively determine if PROs assessing multidimensional palliative domains of care improve the utility of PROs to identify patients with unmet specialty referral needs.
Our findings are overall consistent with previous subgroup analyses of oncology patients, which have shown the formation of low, moderate, and high symptom groups with a mood-predominant phenotype.23,25 No previous studies have examined latent modeling of PROs in a palliative care population. This analysis improves upon these studies and significantly advances the field toward clinical application in several ways. First, this study models symptom data using continuous severity scales rather than “present or not present” cutoffs, which may increase the utility of PRO phenotypes as referral criteria by more accurately characterizing symptom burden. Second, the study design captures a snapshot of a referred patient population before palliative intervention. By defining a clinically relevant study population and time point for PRO collection, our findings can be applied in future validation studies. Third, no previous subgroup analyses have incorporated patient-reported measures of social and existential distress. Brief screening questions for these domains are associated with more severe phenotypes and may provide added utility for screening for palliative needs. Finally, the data set includes a comprehensive set of palliative domain measures collected at multiple clinical sites, which strengthens the validity and generalizability of our findings.
This analysis has limitations. First, the study is cross-sectional, which precludes our ability to validate the phenotypes based on response to palliative care intervention. However, we believe that the PRO phenotypes are clinically relevant given their good construct validity, coupled with the strong evidence base linking unmet symptom needs to adverse patient outcomes.1,3,4,18-20 Second, the sample consists of a large proportion of patients with lung cancer, which could bias the formation of latent subgroups. To address this concern, we conducted a sensitivity analysis restricted to patients with non-lung cancer, which resulted in little change to the model and suggested that influence was small. Third, the proportion of patients in the “Low” phenotype (39%) likely indicates undetected reasons for palliative care referral, such as advanced care planning needs or care coordination, that are not captured within the data set. Fourth, we excluded patients with missing ESAS data, which could result in selection bias toward a healthier population. However, we believe that this bias is small given the large proportion of patients with poor prognosis and performance status included in the study population. Finally, although we did detect statistically significant associations of demographic and clinical characteristics with PRO phenotypes, the precision of these estimates is low because of stratification by latent subgroups. Given the above limitations, the identified PRO phenotypes should be refined and validated in a prospective study to develop PRO-based referral criteria for palliative care.
In summary, our findings suggest that a brief set of PRO measures collected in routine clinical practice can be used to identify PRO phenotypes that characterize physical, psychosocial, and existential distress needs. Future research is needed to validate PRO phenotypes in a patient population before referral. Since the PRO phenotypes are identified among oncology patients referred to palliative care, the phenotypes will likely identify the most high need patients in a general oncology population and will need further refinement to capture earlier palliative care needs. As PRO monitoring grows in clinical practice and larger data sets are available for population-based research,33 latent modeling approaches may have increasing utility to guide supportive interventions to improve health outcomes.
Supported by the Palliative Care Research Cooperative Group funded by the National Institute of Nursing Research (U2CNR014637). T.L.K. was supported by a NIH Cancer Clinical Epidemiology Training Grant (T32CA009679). A.H.K. was supported by an Agency for Healthcare Research and Quality Career Development Award (5K08HS023681-05). The funding sources were not involved in the design and conduct of the study, interpretation of the data, preparation of the manuscript, or in the decision to submit the manuscript for publication.
Conception and design: Tara L. Kaufmann, Jesse Y. Hsu, Arif H. Kamal, Angela DeMichele
Administrative support: Arif H. Kamal
Collection and assembly of data: Tara L. Kaufmann, Kelly D. Getz, Jesse Y. Hsu, Arif H. Kamal, Angela DeMichele
Data analysis and interpretation: All authors
Manuscript writing: All authors
Final approval of manuscript: All authors
Accountable for all aspects of the work: All authors
The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated unless otherwise noted. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or ascopubs.org/op/authors/author-center.
Open Payments is a public database containing information reported by companies about payments made to US-licensed physicians (Open Payments).
Tara L. Kaufmann
Consulting or Advisory Role: PotentiaMetrics
Samuel U. Takvorian
Stock and Other Ownership Interests: Gilead Sciences, Merck, Portola Pharmaceuticals, Bluebird Bio, Johnson & Johnson, Novartis, Pfizer
Consulting or Advisory Role: Health Advances, Novartis, Boehringer Ingelheim, Novocure
Research Funding: Pfizer, AstraZeneca, Lilly, Trizell
Arif H. Kamal
Employment: Prepped Health, Acclivity Health, Votive Health
Leadership: Prepped Health, Acclivity Health, Votive Health
Stock and Other Ownership Interests: Acclivity Health
Consulting or Advisory Role: Medtronic, Huron Therapeutics, New Century Health, Compassus, AstraZeneca, Janssen Oncology, United Health Group
Travel, Accommodations, Expenses: Janssen Oncology
Angela DeMichele
Consulting or Advisory Role: Context Therapeutics, Pfizer
Research Funding: Pfizer, Genentech, Calithera Biosciences, Menarini/Silicon Biosystems
No other potential conflicts of interest were reported.
FIG A1. aThe following answer choices are treated as missing for all ESAS measures and social and existential distress questions: “Unknown”, “Unable to Respond”, and “Other.” ESAS, Edmonton Symptom Assessment System; PRO, patient-reported outcome.
FIG A2. Latent profile analysis results. AIC, Akaike information criterion; BIC, Bayesian information criterion.
FIG A3. Lung cancer senstivity analysis.
|
|
| 1. | Temel JS, Greer JA, Muzikansky A, et al: Early palliative care for patients with metastatic non-small-cell lung cancer. N Engl J Med 363:733-742, 2010 Crossref, Medline, Google Scholar |
| 2. | Temel JS, Greer JA, El-Jawahri A, et al: Effects of early integrated palliative care in patients with lung and GI cancer: A randomized clinical trial. J Clin Oncol 35:834-841, 2017 Link, Google Scholar |
| 3. | Bakitas M, Lyons KD, Hegel MT, et al: Effects of a palliative care intervention on clinical outcomes in patients with advanced cancer: The project ENABLE II randomized controlled trial. JAMA 302:741-749, 2009 Crossref, Medline, Google Scholar |
| 4. | Bakitas MA, Tosteson TD, Li Z, et al: Early versus delayed initiation of concurrent palliative oncology care: Patient outcomes in the ENABLE III randomized controlled trial. J Clin Oncol 33:1438-1445, 2015 Link, Google Scholar |
| 5. | Zimmermann C, Swami N, Krzyzanowska M, et al: Early palliative care for patients with advanced cancer: A cluster-randomised controlled trial. Lancet 383:1721-1730, 2014 Crossref, Medline, Google Scholar |
| 6. | Scibetta C, Kerr K, McGuire J, et al: The costs of waiting: Implications of the timing of palliative care consultation among a cohort of decedents at a comprehensive cancer center. J Palliat Med 19:69-75, 2016 Crossref, Medline, Google Scholar |
| 7. | Kavalieratos D, Corbelli J, Zhang D, et al: Association between palliative care and patient and caregiver outcomes: A systematic review and meta-analysis. JAMA 316:2104-2114, 2016 Crossref, Medline, Google Scholar |
| 8. | Ferrell BR, Temel JS, Temin S, et al: Integration of palliative care into standard oncology care: American society of clinical oncology clinical practice guideline update. J Clin Oncol 35:96-112, 2017 Link, Google Scholar |
| 9. | Hui D, Elsayem A, De la Cruz M, et al: Availability and integration of palliative care at US cancer centers. JAMA 303:1054-1061, 2010 Crossref, Medline, Google Scholar |
| 10. | Courtright KR, Cassel JB, Halpern SD: A research agenda for high-value palliative care. Ann Intern Med 168:71-72, 2018 Crossref, Medline, Google Scholar |
| 11. | Von Roenn JH, Temel J: The integration of palliative care and oncology: The evidence. Oncology (Williston Park) 25:1258-1260, 1262, 1264-1265, 2011 Medline, Google Scholar |
| 12. | Bakitas MA, El-Jawahri A, Farquhar M, et al: The TEAM approach to improving oncology outcomes by incorporating palliative care in practice. J Oncol Pract 13:557-566, 2017 Link, Google Scholar |
| 13. | Hui D, Meng YC, Bruera S, et al: Referral criteria for outpatient palliative cancer care: A systematic review. Oncologist 21:895-901, 2016 Crossref, Medline, Google Scholar |
| 14. | Hui D, Mori M, Watanabe SM, et al: Referral criteria for outpatient specialty palliative cancer care: An international consensus. Lancet Oncol 17:e552-e559, 2016 Crossref, Medline, Google Scholar |
| 15. | Glare PA, Semple D, Stabler SM, et al: Palliative care in the outpatient oncology setting: Evaluation of a practical set of referral criteria. J Oncol Pract 7:366-370, 2011 Link, Google Scholar |
| 16. | Adelson K, Paris J, Horton JR, et al: Standardized criteria for palliative care consultation on a solid tumor oncology service reduces downstream health care use. J Oncol Pract 13:e431-e440, 2017 Link, Google Scholar |
| 17. | Glare PA, Chow K: Validation of a simple screening tool for identifying unmet palliative care needs in patients with cancer. J Oncol Pract 11:e81-e86, 2015 Link, Google Scholar |
| 18. | Basch E, Deal AM, Kris MG, et al: Symptom monitoring with patient-reported outcomes during routine cancer treatment: A randomized controlled trial. J Clin Oncol 34:557-565, 2016 Link, Google Scholar |
| 19. | Basch E, Deal AM, Dueck AC, et al: Overall survival results of a trial assessing patient-reported outcomes for symptom monitoring during routine cancer treatment. JAMA 318:197-198, 2017 Crossref, Medline, Google Scholar |
| 20. | Denis F, Basch E, Septans A-L, et al: Two-year survival comparing web-based symptom monitoring vs routine surveillance following treatment for lung cancer. JAMA 321:306-307, 2019 Crossref, Medline, Google Scholar |
| 21. | Dajani DR, Llabre MM, Nebel MB, et al: Heterogeneity of executive functions among comorbid neurodevelopmental disorders. Sci Rep 6:36566, 2016 Crossref, Medline, Google Scholar |
| 22. | Avis NE, Levine B, Marshall SA, et al: Longitudinal examination of symptom profiles among breast cancer survivors. J Pain Symptom Manage 53:703-710, 2017 Crossref, Medline, Google Scholar |
| 23. | Miaskowski C, Dunn L, Ritchie C, et al: Latent class analysis reveals distinct subgroups of patients based on symptom occurrence and demographic and clinical characteristics. J Pain Symptom Manag 50:28-37, 2015 Crossref, Medline, Google Scholar |
| 24. | Lanza ST, Rhoades BL: Latent class analysis: An alternative perspective on subgroup analysis in prevention and treatment. Prev Sci 14:157-168, 2013 Crossref, Medline, Google Scholar |
| 25. | Astrup GL, Hofsø K, Bjordal K, et al: Patient factors and quality of life outcomes differ among four subgroups of oncology patients based on symptom occurrence. Acta Oncol 56:462-470, 2017 Crossref, Medline, Google Scholar |
| 26. | Papachristou N, Barnaghi P, Cooper BA, et al: Congruence between latent class and K-modes analyses in the identification of oncology patients with distinct symptom experiences. J Pain Symptom Manage 55:318-333.e4, 2018 Crossref, Medline, Google Scholar |
| 27. | Kamal AH, Harrison KL, Bakitas M, et al: Improving the quality of palliative care through national and regional collaboration efforts. Cancer Control 22:396-402, 2015 Crossref, Medline, Google Scholar |
| 28. | Kamal AH, Bull J, Kavalieratos D, et al: Development of the quality data collection tool for prospective quality assessment and reporting in palliative care. J Palliat Med 19:1148-1155, 2016 Crossref, Medline, Google Scholar |
| 29. | Ferrell BR, Twaddle ML, Melnick A, et al: National consensus project clinical practice guidelines for quality palliative care guidelines, 4th edition. J Palliat Med 21:1684-1689, 2018 Crossref, Medline, Google Scholar |
| 30. | Bruera E, Kuehn N, Miller MJ, et al: The Edmonton Symptom Assessment system (ESAS): A simple method for the assessment of palliative care patients. J Palliat Care 7:6-9, 1991 Crossref, Medline, Google Scholar |
| 31. | Kroenke CH, Rosner B, Chen WY, et al: Functional impact of breast cancer by age at diagnosis. J Clin Oncol 22:1849-1856, 2004 Link, Google Scholar |
| 32. | Trevino KM, Maciejewski PK, Fasciano K, et al: Coping and psychological distress in young adults with advanced cancer. J Support Oncol 10:124-130, 2012 Crossref, Medline, Google Scholar |
| 33. | Basch E, Wilfong L, Schrag D: Adding patient-reported outcomes to Medicare's oncology value-based payment model. JAMA 323:213-214, 2020 Crossref, Medline, Google Scholar |
