Symptom-BERT: Enhancing Cancer Symptom Detection in EHR Clinical Notes

CONTEXT

Extracting cancer symptom documentation allows clinicians to develop highly individualized symptom prediction algorithms to deliver symptom management care. Leveraging advanced language models to detect symptom data in clinical narratives can significantly enhance this process.

OBJECTIVE

This study uses a pre-trained large language model to detect and extract cancer symptoms in clinical notes.

METHODS

We developed a pre-trained language model to identify cancer symptoms in clinical notes based on a clinical corpus from the Enterprise Data Warehouse for Research at a healthcare system in the Midwestern United States. This study was conducted in 4 phases:1 pre-training a Bio-Clinical BERT model on 1 million unlabeled clinical documents,2 fine-tuning Symptom-BERT for detecting 13 cancer symptom groups within 1112 annotated clinical notes,3 generating 180 synthetic clinical notes using ChatGPT-4 for external validation, and4 comparing the internal and external performance of Symptom-BERT against a non-pre-trained version and six other BERT implementations.

RESULTS

The Symptom-BERT model effectively detected cancer symptoms in clinical notes. It achieved results with a micro-averaged F1-score of 0.933, an AUC of 0.929 internally, and 0.831 and 0.834 externally. Our analysis shows that physical symptoms, like Pruritus, are typically identified with higher performance than psychological symptoms, such as Anxiety.

CONCLUSION

This study underscores the transformative potential of specialized pre-training on domain-specific data in boosting the performance of language models for medical applications. The Symptom-BERT model's exceptional efficacy in detecting cancer symptoms heralds a groundbreaking stride in patient-centered AI technologies, offering a promising path to elevate symptom management and cultivate superior patient self-care outcomes.

Defining research priorities and needs in cancer symptoms for adults diagnosed with cancer: an Australian/New Zealand modified Delphi study

PURPOSE

This study asked consumers (patients, carers) and healthcare professionals (HCPs) to identify the most important symptoms for adults with cancer and potential treatment interventions.

METHODS

A modified Delphi study was conducted involving two rounds of electronic surveys based on prevalent cancer symptoms identified from the literature. Round 1 gathered information on participant demographics, opinions and/or experience on cancer symptom frequency and impact, and suggestions for interventions and/or service delivery models for further research to improve management of cancer symptoms. In Round 2, respondents ranked the importance of the top ten interventions identified in Round 1. In Round 3, separate expert panels of consumers and healthcare professionals (HCPs) attempted to reach consensus on the symptoms and interventions previously identified.

RESULTS

Consensus was reached for six symptoms across both groups: fatigue, constipation, diarrhoea, incontinence, and difficulty with urination. Notably, fatigue was the only symptom to reach consensus across both groups in Round 1. Similarly, consensus was reached for six interventions across both groups. These were the following: medicinal cannabis, physical activity, psychological therapies, non-opioid interventions for pain, opioids for breathlessness and cough, and other pharmacological interventions.

CONCLUSIONS

Consumers and HCPs prioritise differently; however, the symptoms and interventions that reached consensus provide a basis for future research. Fatigue should be considered a high priority given its prevalence and its influence on other symptoms. The lack of consumer consensus indicates the uniqueness of their experience and the need for a patient-centred approach. Understanding individual consumer experience is important when planning research into better symptom management.

Nurse-led emergency department avoidance model of care for patients receiving cancer therapy in the ambulatory setting: a health service improvement initiative

AIMS

The Symptom and Urgent Review Clinic was a service improvement initiative, which consisted of the implementation and evaluation of a nurse-led emergency department (ED) avoidance model of care. The clinic was developed for patients experiencing symptoms associated with systemic anti-cancer therapy in ambulatory cancer settings.

METHODS

The clinic was implemented in four health services in Melbourne, Australia across a six-month period in 2018. Evaluation was by prospective data collection of the frequency and characteristics of patients who used the service, pre- and post-survey of patient reported experience, and a post-implementation survey of clinician engagement and experience.

RESULTS

There were 3095 patient encounters in the six-month implementation period; 136 patients were directly admitted to inpatient healthcare services after clinic utilization. Of patients who contacted SURC (n = 2174), a quarter (n = 553) stated they would have otherwise presented to the emergency department and 51% (n = 1108) reported they would have otherwise called the Day Oncology Unit. After implementation, more patients reported having a dedicated point of contact (OR 14.3; 95% CI 5.8-37.7) and ease of contacting the nurse (OR 5.5; 95% CI 2.6-12.1). Clinician reported experience and engagement with the clinic was highly favorable.

CONCLUSION

The nurse-led emergency department avoidance model of care addressed a gap in service delivery, while optimizing service utilization by reducing ED presentations. Patients reported improved levels of satisfaction with ease of access to a dedicated nurse and advice provided.

Acupuncture in hematologic malignancies and hematopoietic cell transplantation

Acupuncture is an integrative medicine modality that has been used historically and studied recently for managing various symptoms and diseases. Patients with hematological malignancies often experience a high symptom burden including pain, fatigue, nausea, vomiting, and insomnia. Pharmacological interventions are the mainstay of managing these symptoms, although they may not provide adequate control or are not well tolerated due to side effects. There have been many randomized controlled trials of acupuncture on reduction of these symptoms. In this review, we will summarize the current evidence on the effect of acupuncture on the management of common symptoms experienced by patients with hematological malignancies and the current understanding of acupuncture's mechanism of action. The evidence is the strongest for reducing pain and nausea vomiting, less strong for symptoms such as peripheral neuropathy, fatigue, constipation, insomnia, night sweat, and pruritus.

Identification of Uncontrolled Symptoms in Cancer Patients Using Natural Language Processing

CONTEXT

For patients with cancer, uncontrolled pain and other symptoms are the leading cause of unplanned hospitalizations. Early access to specialty palliative care (PC) is effective to reduce symptom burden, but more efficient approaches are needed for rapid identification and referral. Information on symptom burden largely exists in free-text notes, limiting its utility as a trigger for best practice alerts or automated referrals.

OBJECTIVES

To evaluate whether natural language processing (NLP) can be used to identify uncontrolled symptoms (pain, dyspnea, or nausea/vomiting) in the electronic health record (EHR) among hospitalized cancer patients with advanced disease.

METHODS

The dataset included 1,644 hospitalization encounters for cancer patients admitted from 1/2017 -6/2019. We randomly sampled 296 encounters, which included 15,580 clinical notes. We manually reviewed the notes and recorded symptom severity. The primary endpoint was an indicator for whether a symptom was labeled as "controlled" (none, mild, not reported) or as "uncontrolled" (moderate or severe). We randomly split the data into training and test sets and used the Random Forest algorithm to evaluate final model performance.

RESULTS

Our models predicted presence of an uncontrolled symptom with the following performance: pain with 61% accuracy, 69% sensitivity, and 46% specificity (F1: 69.5); nausea/vomiting with 68% accuracy, 21% sensitivity, and 90% specificity (F1: 29.4); and dyspnea with 80% accuracy, 22% sensitivity, and 88% specificity (F1: 21.1).

CONCLUSION

This study demonstrated initial feasibility of using NLP to identify hospitalized cancer patients with uncontrolled symptoms. Further model development is needed before these algorithms could be implemented to trigger early access to PC.

Frequency and characteristics of advanced cancer patients with COVID + ve status among inpatient supportive care consults during the pandemic: experience from a tertiary cancer center

Purpose

There is limited literature available regards the frequency and characteristics of COVID-19 + ve status among advanced cancer patients referred to an inpatient supportive care consultation(PC) at a tertiary cancer center. Our study aimed to determine the frequency and characteristics of COVID-19 + ve cancer patients seen by PC.

Methods

Advanced cancer patients seen as a consult by PC between June 15 and September 25, 2020, at MD Anderson Cancer Center were eligible for the study. We evaluated the patient demographics, clinical characteristics including symptoms(ESAS), delirium(MDAS), COVID + status prior to, and after PC referral(converters), and type of PC delivery(in person or virtual care).

Results

Sixty-six out of 1380 (4.8%) PC consults were COVID-19 + ve: 42 prior to PC (79%), and 14 (21%) were COVID-19 + ve after the PC (converters). COVID-19 + PC patients had lower depression (P = .035), spiritual distress (P = .003), and were more seen frequently virtually (P < 0.001). There was no significant difference between COVID-19-ve patients and converters. Converters had higher symptom distress (P = 0.007), lower delirium (P = 0.014), and were referred earlier (P = .011) compared to COVID + PC patients diagnosed prior to PC consult. Overall, patients seen virtually compared in-person by PC were younger (P = 0.02) and had lower delirium (P = 0.007).

Conclusion

The burden of COVID-19 + status among patients referred to PC was low. COVID-19 + ve patients had more frequent virtual visits, lower depression, and spiritual distress scores. Patient seen virtually were significantly younger and had lower delirium. During a new pandemic, universal virtual care might be emphasized especially at initial encounters after admission and further research is needed on the potential efficacy of this intervention.

Supplementary Information

The online version contains supplementary material available at 10.1007/s00520-021-06525-x.

Awareness of Common Cancer Risk Factors and Symptoms in Saudi Arabia: A Community-Based Study

Objective:

To assess the level of cancer awareness and the relationship between the awareness of common cancer symptoms and risk factors and the sample’s sociodemographic profile.

Methods:

A community-based cross-sectional study conducted in Tabuk city, a convenient sample of 675 participants completed a questionnaire about common factors causing cancer and symptoms related to cancer. Descriptive statistics and chi-squared analysis were used to measure sample characteristics and their association with knowledge of cancer risk factors and symptoms.

Results:

Level of education and family history were significantly related to awareness of cancer risk factors (P= 0.017) and (P= 0.048), respectively. Factors were significantly associated with awareness of cancer symptoms include Gender (P=0.000), nationality (P=0.013), and undergoing regular cancer screening tests (0.008). Internet was the primary source of information about cancer and related significantly to knowledge about cancer symptoms(P=0.000) and risk factors(P=0.00). More than half of the sample scored poorly for knowledge of both cancer risk factors (58.7%) and symptoms (66.2%) in terms of the overall level of cancer awareness. Smoking and heredity were the most identifiable cancer risk factors, and unexplained pain and weight loss were identified by the majority of study participants.

Conclusion:

The level of community knowledge requires further investigation, and more cancer awareness programs need to be conducted. People could be encouraged to use reliable sources of information to obtain accurate cancer information.

Time from Self-Detection of Symptoms to Seeking Definitive Care among Cervical Cancer Patients

Background:

India had the burden of 97,000 new cases of cervical cancer with 60,000 deaths accounting nearly one-third of global cervical cancer deaths during the year 2018. Cervical cancer is the leading cause of cancer mortality in India. The present study aims to estimate the time interval between self-detection of cervical cancer symptoms and seeking care and different barriers for the possible time lag in seeking care.

Methods:

A cross-sectional study was undertaken from April 2017 to September 2017 in a regional cancer centre in the south of India. The centre has both a population and a hospital-based cancer registry. Cervical cancer cases (N= 210) with histological confirmation were interviewed at the hospital using a pre-tested semi-structured questionnaire.

Results:

The median time interval between the self-detection of cervical cancer symptoms and first contact with the general physician was 80 [IQR 45-150] days. The overall median time interval between the self-detection of symptoms to the initiation of primary treatment was 123[IQR 83-205] days. The major perceived reason for not seeking medical care was a lack of awareness in identifying cervical cancer symptoms in 183(92.9%) women.

Conclusion:

The median time of 80 days was observed from the self-detection of cervical cancer symptoms to the first contact with a general physician. Lack of awareness of patients pertaining to cancer symptoms was the major concern in seeking cancer care.

The impact of a volunteer-led community cancer awareness programme on knowledge of cancer risk factors and symptoms, screening, and barriers to seeking help

OBJECTIVE

To analyse the impact of a community cancer awareness programme on knowledge of cancer risk factors and symptoms, screening, and barriers to seeking help.

METHODS

Personalised information through peer-led champions was delivered to 5500 people in a range of settings and Cancer Awareness Measures questionnaires were completed by 119 participants at pre-arranged sessions (convenience sampling) before and after the intervention. Data were analysed using McNemar tests, Mann-Whitney U test and a Wilcoxon Signed Rank test.

RESULTS

Data showed increase in knowledge after the intervention for cancer screening programmes (p < 0.05), recognition of warning signs for cancer (p < 0.05), and recognition of risk factors for cancer in seven of the eleven options (p < 0.001). Results suggest a decrease in perception of barriers to seeking help (p < 0.05). The intervention had a stronger impact on recognition of cancer symptoms for people who have been affected by cancer (p = 0.02).

CONCLUSION

The Cancer Awareness Measures questionnaire proved an effective tool for evaluation and awareness improved after the intervention amongst those who completed it.

PRACTICE IMPLICATIONS

Enhancing the perceived personal relevance of information to those with experience of cancer may improve information processing and retention. The study highlights cancer awareness gaps among the public for future intervention development.

Validation of self-reported help-seeking, and measurement of the patient interval, for cancer symptoms: an observational study to inform methodological challenges in symptomatic presentation research

BACKGROUND

To improve earlier presentation with potential symptoms of cancer, accurate data are needed on how people respond to these symptoms. It is currently unclear how self-reported medical help-seeking for symptoms associated with cancer by people from the community correspond to what is recorded in their general practice records, or how well the patient interval (time from symptom onset to first presentation to a health-professional) can be estimated from patient records.

METHOD

Data from two studies that reviewed general practice electronic records of residents in Scotland, (i) the 'Useful Study': respondents to a general population survey who reported experiencing symptoms potentially associated with one of four common cancers (breast, colorectal, lung and upper gastro-intestinal) and (ii) the 'Detect Cancer Early' programme: cancer patients with one of the same four cancers. Survey respondents' self-reported help-seeking (yes/no) was corroborated; Cohen's Kappa assessed level of agreement. Combined data on the patient interval were evaluated using descriptive analysis.

RESULTS

'Useful Study' respondents' self-report of help-seeking showed exact correspondence with general practice electronic records in 72% of cases (n = 136, kappa 0.453, moderate agreement). Between both studies, 1269 patient records from 35 general practices were reviewed. The patient interval could not be determined in 44% (n = 809) of symptoms presented by these individuals.

CONCLUSIONS

Patient self-report of help-seeking for symptoms potentially associated with cancer offer a reasonably accurate method to research responses to these symptoms. Incomplete patient interval data suggest routine general practice records are unreliable for measuring this important part of the patient's symptom journey.

Quality of Life in Patients with Advanced Non-Small-Cell Lung Cancer Receiving Palliative Chemotherapy

The goal of this study was to explore quality of life in patients with advanced non-small-cell lung cancer (NSCLC) in an attempt to single out features that could help predict the possibility of non-completion of chemotherapy. The survey tool was the Quality of Life Questionnaire Core-30 (QLQ-C30) with the module Lung Cancer 13 (LC-13) developed by the European Organization for Research and Treatment of Cancer. The assessment of quality of life (QoL) was performed in 58 patients with advanced NSCLC before palliative chemotherapy and it was repeated in 43 patients who completed at least three cycles of chemotherapy. We found that the patients who failed to complete the chemotherapy course distinctly showed, in contradistinction to those who completed it, poor physical functioning in (67.6 ± 16.3 vs. 78.3 ± 21.3 points, respectively, p < 0.05) and the lack of appetite (27.1 ± 38.0 vs. 48.9 ± 37.5 points, respectively p < 0.05). At the end of palliative chemotherapy alopecia, sore throat, and constipation significantly worsened QoL, but global health status remained unchanged. In conclusion, poor physical functioning and loss of appetite seem to harbinger a risk of non-completion of chemotherapy in advanced NSCLC.

Evidence, safety and recommendations for when to use acupuncture for treating cancer related symptoms: a narrative review

Background

Recently research on acupuncture for cancer related symptoms has significantly increased. To what extent have medical professionals recommended to use acupuncture in light of that evidence?

Methods

Evidence of effectiveness and safety was found by searching Pubmed for reviews to identify for what conditions and general results. Publications that recommend acupuncture in oncology were searched in the database of an ongoing general search for publications that recommend acupuncture. This database was developed by searching google for publications that recommend the use of acupuncture with the terms ‘name of symptom’ and ‘clinical practice guideline’ or ‘treatment guideline’.

Results

Acupuncture is moderately or weakly effective for 19 symptoms in patients with cancer and cancer survivors. Acupuncture is a safe therapy in cancer care if administered by trained acupuncturists. Acupuncture is targeted to improve symptoms associated with the cancer and different cancer treatments, not to treat the cancer itself. More than 350 publications by clinical practice guideline groups and expert groups, including public health statements made by national and government agencies recommended the use of acupuncture for 61 cancer related symptoms many with positive evidence of effectiveness.

Conclusion

The strength of evidence is weak for many indications, however the evidence for many standard therapies is either not very strong or if stronger, the incidence of adverse events is more, which makes acupuncture a treatment option despite the weak evidence. We have found evidence that many oncologists around the world have started to incorporate acupuncture into the treatment of various cancer related symptoms.

Utilizing the Patient Reported Outcomes Measurement Information System (PROMIS®) to increase referral to ancillary support services for severely symptomatic patients with gynecologic cancer

OBJECTIVE

The Patient-Reported Outcomes Measurement Information System (PROMIS®) Network has developed a comprehensive repository of electronic patient reported outcomes measures (ePROs) of major symptom domains that have been validated in cancer patients. Their use for patients with gynecologic cancer has been understudied. Our objective was to establish feasibility and acceptability of PROMIS ePRO integration in a gynecologic oncology outpatient clinic and assess if it can help identify severely symptomatic patients and increase referral to supportive services.

METHODS

English-speaking patients with a confirmed history of gynecologic cancer completed PROMIS ePROs on iPads in the waiting area of an outpatient gynecologic oncology clinic. Symptom scores were calculated for each respondent and grouped using documented severity thresholds. Response data was compared with clinicopathologic characteristics across symptom domains. Severely symptomatic patients were offered referral to ancillary services and asked to complete post-exposure surveys assessing acceptability of the ePRO.

RESULTS

Of the 336 patients who completed ePROs, 35% had active disease and 19% had experienced at least one disease recurrence. Sixty-nine percent of the cohort demonstrated moderate to severe physical dysfunction (60%), pain (36%), fatigue (28%), anxiety (9%), depression (8%), and sexual dysfunction (32%). Thirty-nine (12%) severely symptomatic patients were referred to services such as psychiatry, palliative care, pain management, social work or integrative oncology care. Most survey respondents identified the ePROs as helpful (78%) and easy to complete (92%).

CONCLUSIONS

Outpatient PROMIS ePRO administration is feasible and acceptable to gynecologic oncology patients and can help identify severely symptomatic patients for referral to ancillary support services.

Understanding the roles of patient symptoms and subjective appraisals in well-being among breast cancer patients

PURPOSE

To examine the roles of both patient symptoms, and subjective appraisals of stress (self-efficacy, symptom barriers, symptom distress), in understanding well-being (anxiety, depression, cancer-specific quality of life, mental health quality of life, and physical health quality of life) in breast cancer patients.

METHODS

We examined data from 104 female breast cancer patients. Using a stress process model, we hypothesized that while high levels of patient symptoms would be associated with poorer patient well-being, these effects would be mediated by subjective appraisals, including patient self-efficacy, perceived symptom barriers, and symptom distress.

RESULTS

As expected, higher levels of patient symptoms were associated with poorer well-being on all five indicators. Subjective appraisals of stress added significantly to predictors of well-being, and were mediators of this relationship across all five outcomes.

CONCLUSIONS

While patient symptoms are important predictors of patient well-being, subjective appraisals of the stressfulness of symptoms, and of patients' self-efficacy in managing symptoms, are also key factors. The findings suggest the utility of a stress process model in understanding well-being in breast cancer patients, and point to the potential value of targeting patient appraisals as well as symptoms to improve psychological well-being and quality of life.

From symptom discovery to treatment - women's pathways to breast cancer care: a cross-sectional study

Background

Typically, women in South Africa (SA) are diagnosed with breast cancer when they self-present with symptoms to health facilities. The aim of this study was to determine the pathway that women follow to breast cancer care and factors associated with this journey.

Methods

A cross-sectional study was conducted at a tertiary hospital in the Western Cape Province, SA, between May 2015 and May 2016. Newly diagnosed breast cancer patients were interviewed to determine their socio-demographic profile; knowledge of risk factors, signs and symptoms; appraisal of breast changes; clinical profile and; key time events in the journey to care. The Model of Pathways to Treatment Framework underpinned the analysis. The total time (TT) between a woman noticing the first breast change and the date of scheduled treatment was divided into 3 intervals: the patient interval (PI); the diagnostic interval (DI) and the pre-treatment interval (PTI). For the PI, DI and PTI a bivariate comparison of median time intervals by various characteristics was conducted using Wilcoxon rank-sum and Kruskal-Wallis tests. Cox Proportional-Hazards models were used to identify factors independently associated with the PI, DI and PTI.

Results

The median age of the 201 participants was 54 years, and 22% presented with late stage disease. The median TT was 110 days, with median patient, diagnostic and pre-treatment intervals of 23, 28 and 37 days respectively. Factors associated with the PI were: older age (Hazard ratio (HR) 0.59, 95% CI 0.40–0.86), initial symptom denial (HR 0.43, 95% CI 0.19–0.97) and waiting for a lump to increase in size before seeking care (HR 0.51, 95% CI 0.33–0.77). Women with co-morbidities had a significantly longer DI (HR 0.67, 95% CI 0.47–0.96) as did women who mentioned denial of initial breast symptoms (HR 4.61, 95% CI 1.80–11.78). The PTI was associated with late stage disease at presentation (HR 1.78, 95% CI 1.15–2.76).

Conclusion

The Model of Pathways to Treatment provides a useful framework to explore patient’s journeys to care and identified opportunities for targeted interventions.

Electronic supplementary material

The online version of this article (10.1186/s12885-018-4219-7) contains supplementary material, which is available to authorized users.

Efficacy and Safety of the Traditional Herbal Medicine, Gamiguibi-tang, in Patients With Cancer-Related Sleep Disturbance: A Prospective, Randomized, Wait-List-Controlled, Pilot Study

Background: Sleep disturbance is the second most bothersome symptom in patients with cancer, and it can significantly impair their quality of life. The aim of this study was to investigate the efficacy and safety of the traditional herbal medicine Gamiguibi-tang (GGBT) in patients with cancer-related sleep disturbance. Methods: We conducted a prospective, randomized, wait-list-controlled, open-label pilot clinical trial on cancer-related sleep disturbance. Patients with cancer experiencing poor sleep quality with a Pittsburgh Sleep Quality Index of at least 6 were randomly assigned to the GGBT and wait-list groups to receive GGBT and conventional care, respectively, for 2 weeks. The primary endpoint was the Insomnia Severity Index (ISI) score. Fatigue, depression, and cognitive impairment were assessed as the secondary endpoints by using the Brief Fatigue Inventory (BFI), Beck Depression Inventory (BDI), and Montreal Cognitive Assessment (MoCA). Results: Thirty participants who met the eligibility criteria were enrolled. Sleep disturbance assessed using the ISI improved significantly more in the GGBT group than in the wait-list group (−5.5 ± 4.4 vs 0.1 ± 1.1, P < .001). Fatigue level determined using the BFI also improved significantly more in the GGBT group than in the wait-list group (−0.8 ± 0.8 vs 0.0 ± 0.3, P = .002). The BDI and MoCA scores showed no significant changes. Adverse events were reported in two patients in the GGBT group and consisted of mild dyspepsia and mild edema. Conclusion: GGBT may be a potential treatment option for cancer-related sleep disturbance. Further research is needed to investigate the efficacy and safety of GGBT.

Longitudinal study on the impact of physical activity on the symptoms of lung cancer survivors

PURPOSE

To examine the effect of physical activity on the physical and psychosocial symptoms of lung cancer survivors.

METHODS

A longitudinal design was used in this study. Participants were recruited from the chest and surgical departments of medical centers in Taiwan. The instruments used were the Godin Leisure-Time Exercise Questionnaire and the Taiwanese version of the M.D. Anderson Symptom Inventory.

RESULTS

In total, 185 survivors were followed up for 6 months (response rate 66%). Disturbed sleep was the most prevalent symptom in the participants. A generalized estimating equation (GEE) method was employed to analyze the relationships among intensity of physical activity, symptom severity, and symptom interference in the daily life of the participants. Regarding symptom severity, significant differences were observed in fatigue, drowsiness, and disturbed sleep between the participants who engaged in moderate physical activity and those who did not engage in any physical activity. Regarding symptom interference, the participants who engaged in light physical activity experienced a significantly lower level of symptom interference than did those with a sedentary lifestyle.

CONCLUSION

This is the first study to explore the role of physical activity in alleviating symptoms in lung cancer survivors by using the GEE method. The results suggest that physical activity plays an essential role in alleviating the physical and psychological symptoms of lung cancer survivors.