AI-Driven Prediction of Cancer Pain Episodes: A Hybrid Decision Support Approach

Kunzhong Zhou; Li Zhuang; Philip Leung-Ho Yu; Tingting Song; Weifang Wang; Yifeng Guo; Yipeng Zhuang; Yuewen Li; Yuheng Wu; Zhiyong Wang

arxiv: 2512.16739 · v2 · pith:VAY3II32new · submitted 2025-12-18 · 💻 cs.AI

AI-Driven Prediction of Cancer Pain Episodes: A Hybrid Decision Support Approach

Yipeng Zhuang , Yifeng Guo , Yuewen Li , Yuheng Wu , Philip Leung-Ho Yu , Tingting Song , Zhiyong Wang , Kunzhong Zhou

show 2 more authors

Weifang Wang Li Zhuang

This is my paper

Pith reviewed 2026-05-22 12:56 UTC · model grok-4.3

classification 💻 cs.AI

keywords cancer pain predictionhybrid machine learninglarge language modelsbreakthrough painelectronic health recordslung cancerpain episode forecastingdecision support system

0 comments

The pith

A hybrid machine learning and large language model pipeline predicts breakthrough pain episodes in lung cancer patients 48 to 72 hours ahead.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper aims to show that combining machine learning to track temporal trends in medication use with a large language model to interpret unclear dosing records and free-text notes leads to more accurate forecasts of pain episodes. This hybrid method was tested on records from 266 hospitalized lung cancer patients. If effective, it could support earlier interventions for the many patients who experience breakthrough pain. The results show higher accuracy and sensitivity compared to using machine learning by itself.

Core claim

The hybrid approach integrates a machine learning module that captures temporal medication trends with a large language model that interprets ambiguous dosing records and free-text clinical notes, resulting in improved prediction of pain episodes within 48 and 72 hours of hospitalization, with accuracies of 0.876 and 0.917 on a retrospective cohort of 266 inpatients and sensitivity improvements of 10.6% and 10.7% over machine learning alone.

What carries the argument

The hybrid pipeline using machine learning for structured temporal trends and large language models for unstructured clinical data interpretation.

If this is right

Provides a clinically interpretable tool for early pain episode forecasting in oncology.
Has potential to enhance treatment precision and optimize resource allocation.
Offers a scalable approach that works with both structured and unstructured electronic health record data.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

Prospective validation studies could test whether the gains persist when documentation practices vary in real time.
The method might extend to predicting pain in other cancer types or chronic conditions with similar data sources.
Reducing reliance on retrospective notes through direct patient input could address potential biases in the current setup.

Load-bearing premise

The retrospective cohort of 266 inpatients and the large language model's interpretations of unstructured notes accurately reflect true pain episodes and medication patterns without major documentation bias.

What would settle it

A prospective clinical trial comparing the hybrid model's real-time predictions against actual patient pain reports or required interventions, to check whether the 10 percent sensitivity gains over machine learning alone hold up.

read the original abstract

Lung cancer patients frequently experience breakthrough pain episodes, with up to 91% requiring timely intervention. To enable proactive pain management, we propose a hybrid machine learning and large language model pipeline that predicts pain episodes within 48 and 72 hours of hospitalization using both structured and unstructured electronic health record data. A retrospective cohort of 266 inpatients was analyzed, with features including demographics, tumor stage, vital signs, and WHO-tiered analgesic use. The machine learning module captured temporal medication trends, while the large language model interpreted ambiguous dosing records and free-text clinical notes. Integrating these modalities improved sensitivity and interpretability. Our framework achieved an accuracy of 0.876 (48h) and 0.917 (72h), with improvements in sensitivity of 10.6% and 10.7%, respectively, attributable to large language model augmentation. This hybrid approach offers a clinically interpretable and scalable tool for early pain episode forecasting, with potential to enhance treatment precision and optimize resource allocation in oncology care.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

Hybrid ML-LLM setup gives a 10% sensitivity lift for 48/72-hour pain prediction on 266 retrospective lung cancer cases, but the gain may trace to label extraction from the same notes rather than independent signals.

read the letter

The paper combines standard machine learning on structured EHR fields like vitals and medication trends with an LLM that parses free-text notes and ambiguous dosing entries to forecast breakthrough pain episodes in lung cancer inpatients. On their 266-patient retrospective cohort it reports accuracies of 0.876 at 48 hours and 0.917 at 72 hours, plus sensitivity gains of roughly 10.6% and 10.7% over the ML baseline alone. That integration is the concrete step forward here, and the authors make a reasonable case that the LLM adds value on the unstructured parts of the record that pure tabular models miss. The focus on a common, distressing symptom and the use of both data types is a practical extension of existing techniques rather than a new theoretical framework. The numbers are specific enough to be checked against the methods section. The central soft spot is the labeling process. If the binary pain-episode outcomes were pulled from the same clinical notes and dosing records that the LLM interprets, the sensitivity improvement could simply reflect better capture of the annotation style instead of new physiological information. The abstract gives no description of how ground truth was adjudicated, whether raters were blinded, or how label extraction was kept separate from feature construction. With a modest cohort size and a retrospective design, that leaves the reported delta hard to interpret without those details. Cross-validation and statistical testing of the improvement are also not mentioned in the summary. This is the sort of applied work that oncology decision-support groups or clinical informatics teams would look at for ideas on multimodal symptom forecasting. A reader already working on EHR-based prediction models would get usable setup details and a clear baseline comparison. It is worth sending to peer review so the label-generation steps and any external validation plans can be examined directly.

Referee Report

3 major / 2 minor

Summary. The paper proposes a hybrid ML-LLM pipeline to predict breakthrough pain episodes in lung cancer inpatients within 48h and 72h windows using structured EHR features (demographics, vitals, WHO-tiered analgesics, temporal medication trends) plus LLM interpretation of ambiguous dosing records and free-text notes. On a retrospective 266-patient cohort it reports accuracies of 0.876 (48h) and 0.917 (72h) together with 10.6% and 10.7% sensitivity gains over an ML-only baseline, attributing the lift to the LLM module.

Significance. If the reported sensitivity gains are shown to arise from genuine augmentation rather than label leakage or annotation-style fitting, the hybrid framework could supply a clinically interpretable early-warning tool for oncology pain management, improving proactive intervention and resource use. The modest cohort size and retrospective design limit immediate generalizability but supply a useful proof-of-concept for multimodal EHR modeling.

major comments (3)

[Abstract / Methods] Abstract and Methods: The headline sensitivity improvements (10.6% at 48h, 10.7% at 72h) are only interpretable if the binary outcome labels (pain episode within the prediction window) are defined independently of the free-text notes and dosing records that the LLM parses. The manuscript supplies no description of label extraction, adjudication protocol, inter-rater reliability, or separation between feature construction and outcome definition; without this the observed delta cannot be confidently attributed to physiological signal rather than reduced label noise.
[Results] Results: No cross-validation procedure, statistical test of the sensitivity delta, or handling of missing data is reported for the 266-patient cohort. These omissions leave the robustness of the accuracy figures (0.876 / 0.917) and the claimed 10% improvement unassessable.
[Methods] Methods: The ML baseline for temporal medication trends and the precise prompting/interpretation rules for the LLM module are not specified in sufficient detail to allow replication or to rule out that the hybrid gain simply reflects better fitting to the same documentation patterns used for labeling.

minor comments (2)

[Abstract] The abstract states that the hybrid model 'improved sensitivity and interpretability' but provides no quantitative metric or example for the interpretability gain.
[Discussion] Potential documentation bias inherent to retrospective EHR extraction should be listed explicitly among the limitations.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for their thoughtful and constructive comments, which have highlighted key areas where additional transparency will strengthen the manuscript. We respond to each major comment below and commit to revisions that address the concerns raised while preserving the integrity of our reported findings.

read point-by-point responses

Referee: [Abstract / Methods] Abstract and Methods: The headline sensitivity improvements (10.6% at 48h, 10.7% at 72h) are only interpretable if the binary outcome labels (pain episode within the prediction window) are defined independently of the free-text notes and dosing records that the LLM parses. The manuscript supplies no description of label extraction, adjudication protocol, inter-rater reliability, or separation between feature construction and outcome definition; without this the observed delta cannot be confidently attributed to physiological signal rather than reduced label noise.

Authors: We agree that a clear account of label extraction is essential for interpreting the sensitivity gains. In the revised Methods section we will add a dedicated subsection describing how binary pain-episode labels were obtained exclusively from structured EHR fields that record clinician-documented breakthrough pain events and standardized pain-assessment scores. These fields are distinct from the free-text notes and dosing records processed by the LLM. We will also report the adjudication protocol (two independent clinical reviewers) together with inter-rater reliability and will explicitly state that label definition occurred prior to and independently of LLM feature construction. This addition will allow readers to confirm that the observed improvements reflect genuine augmentation rather than reduced label noise. revision: yes
Referee: [Results] Results: No cross-validation procedure, statistical test of the sensitivity delta, or handling of missing data is reported for the 266-patient cohort. These omissions leave the robustness of the accuracy figures (0.876 / 0.917) and the claimed 10% improvement unassessable.

Authors: We acknowledge that these details were omitted from the original submission and limit evaluation of robustness. In the revision we will report a stratified train-test split that respects temporal ordering, add k-fold cross-validation results, include a statistical test (McNemar’s test) for the sensitivity delta with associated p-values and confidence intervals, and fully document missing-data handling (multiple imputation by chained equations) together with complete-case sensitivity analyses. These changes will make the performance claims directly assessable. revision: yes
Referee: [Methods] Methods: The ML baseline for temporal medication trends and the precise prompting/interpretation rules for the LLM module are not specified in sufficient detail to allow replication or to rule out that the hybrid gain simply reflects better fitting to the same documentation patterns used for labeling.

Authors: We recognize the need for greater methodological detail to support replication and to address concerns about potential circular fitting. The revised Methods section will specify the exact ML baseline architecture and hyperparameters used to model temporal medication trends, list all engineered features, and provide the full prompt templates together with the deterministic interpretation rules that convert LLM outputs into structured features. These additions will enable independent replication and help demonstrate that performance gains arise from clinically meaningful signal rather than artifactual alignment with labeling patterns. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical results on retrospective EHR cohort are self-contained

full rationale

The paper reports an empirical hybrid ML-LLM pipeline evaluated on a 266-patient retrospective cohort using structured EHR features and unstructured notes. No mathematical derivation chain, equations, or first-principles claims are present that reduce any prediction to a fitted parameter or self-referential quantity by construction. The reported sensitivity gains and accuracies (0.876/0.917) are performance metrics on observed data rather than outputs forced by the model's own definitions or prior self-citations. The pipeline is externally falsifiable via prospective validation and does not invoke uniqueness theorems or ansatzes from the authors' prior work.

Axiom & Free-Parameter Ledger

2 free parameters · 1 axioms · 0 invented entities

The central performance claims rest on the assumption that retrospective EHR data faithfully records pain episodes and that LLM augmentation adds genuine signal rather than noise or overfitting; no free parameters are explicitly named but implicit model fitting and prompting choices are required.

free parameters (2)

ML model hyperparameters and decision thresholds
Fitted during training to optimize accuracy and sensitivity on the 266-patient cohort.
LLM prompting strategy and interpretation rules
Chosen to extract meaning from ambiguous dosing and free-text notes.

axioms (1)

domain assumption Electronic health records contain accurate and complete information on pain episodes, vital signs, and analgesic use.
Invoked implicitly for the retrospective cohort analysis to serve as ground truth.

pith-pipeline@v0.9.0 · 5738 in / 1359 out tokens · 53263 ms · 2026-05-22T12:56:21.665795+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

20 extracted references · 20 canonical work pages · 1 internal anchor

[1]

Pain management in lung cancer,

F. Nurwidya, E. Syahruddin, and F. Yunus, “Pain management in lung cancer,”Advances in Respiratory Medicine, vol. 84, no. 6, pp. 331–336

work page
[2]

Global burden of lung cancer: Prognosis, symptom management, and challenges with pulmonary fibrosis,

D. A. Mohan Jha, D. A. Kumar, D. J. Abraham, and Huma Firdaus, “Global burden of lung cancer: Prognosis, symptom management, and challenges with pulmonary fibrosis,”International Journal of Trends in OncoScience

work page
[3]

Clinical management of pain in advanced lung cancer,

C. P. Simmons, N. Macleod, and B. J. Laird, “Clinical management of pain in advanced lung cancer,”Clinical Medicine Insights: Oncology, vol. 6, p. CMO.S8360

work page
[4]

Cancer pain – adult

MD Anderson Cancer Center, “Cancer pain – adult.”

work page
[5]

Pain experienced by patients with ter- minal head and neck carcinoma,

Y . P. Talmi, A. Waller, M. Bercovici, Z. Horowitz, M. R. Pfeffer, A. Adunski, and J. Kronenberg, “Pain experienced by patients with ter- minal head and neck carcinoma,”Cancer: Interdisciplinary International Journal of the American Cancer Society, vol. 80, no. 6, pp. 1117–1123

work page
[6]

Opioid rotation for cancer pain: rationale and clinical aspects,

S. Mercadante, “Opioid rotation for cancer pain: rationale and clinical aspects,”Cancer: Interdisciplinary International Journal of the Ameri- can Cancer Society, vol. 86, no. 9, pp. 1856–1866

work page
[7]

Prediction of healthcare utilization following an episode of physical therapy for musculoskeletal pain,

T. A. Lentz, J. M. Beneciuk, and S. Z. George, “Prediction of healthcare utilization following an episode of physical therapy for musculoskeletal pain,”BMC Health Services Research, vol. 18, no. 1, p. 648

work page
[8]

Cluster-then-classify methodology for the identification of pain episodes in chronic diseases,

J. Galvez-Goicurla, J. Pagan, A. B. Gago-Veiga, J. M. Moya, and J. L. Ayala, “Cluster-then-classify methodology for the identification of pain episodes in chronic diseases,”IEEE Journal of Biomedical and Health Informatics, vol. 26, no. 5, pp. 2339–2350

work page
[9]

Predictive modeling for end-of-life pain outcome using electronic health records,

M. K. Lodhi, J. Stifter, Y . Yao, R. Ansari, G. M. Keenan, D. J. Wilkie, and A. A. Khokhar, “Predictive modeling for end-of-life pain outcome using electronic health records,” inAdvances in Data Mining: Appli- cations and Theoretical Aspects: 15th Industrial Conference, ICDM 2015, Hamburg, Germany, July 11-24, 2015, Proceedings 15, pp. 56–68, Springer, 2015

work page 2015
[10]

Predicting the risk of cancer in adults using supervised machine learning: a scoping review,

A. A. Alfayez, H. Kunz, and A. G. Lai, “Predicting the risk of cancer in adults using supervised machine learning: a scoping review,”BMJ open, vol. 11, no. 9, p. e047755

work page
[11]

Machine learning approaches to predict symptoms in people with cancer: Systematic review,

N. Zeinali, N. Youn, A. Albashayreh, W. Fan, and S. Gilbertson White, “Machine learning approaches to predict symptoms in people with cancer: Systematic review,”JMIR cancer, vol. 10, p. e52322

work page
[12]

Clinical relevance of deep learning models in predicting the onset timing of cancer pain exacerbation,

Y . H. Bang, Y . H. Choi, M. Park, S.-Y . Shin, and S. J. Kim, “Clinical relevance of deep learning models in predicting the onset timing of cancer pain exacerbation,”Scientific Reports, vol. 13, no. 1, p. 11501

work page
[13]

Predic- tion of cancer symptom trajectory using longitudinal electronic health record data and long short-term memory neural network,

S. Chae, W. N. Street, N. Ramaraju, and S. Gilbertson-White, “Predic- tion of cancer symptom trajectory using longitudinal electronic health record data and long short-term memory neural network,”JCO clinical cancer informatics, vol. 8, p. e2300039

work page
[14]

Large language models in medicine,

A. J. Thirunavukarasu, D. S. J. Ting, K. Elangovan, L. Gutierrez, T. F. Tan, and D. S. W. Ting, “Large language models in medicine,”Nature medicine, vol. 29, no. 8, pp. 1930–1940, 2023

work page 1930
[15]

Table meets llm: Can large language models understand structured table data? a benchmark and empirical study,

Y . Sui, M. Zhou, M. Zhou, S. Han, and D. Zhang, “Table meets llm: Can large language models understand structured table data? a benchmark and empirical study,” inProceedings of the 17th ACM International Conference on Web Search and Data Mining, pp. 645–654, 2024

work page 2024
[16]

Small models are llm knowledge triggers for medical tabular prediction,

J. Yan, J. Chen, C. Hu, B. Zheng, Y . Hu, J. Sun, and J. Wu, “Small models are llm knowledge triggers for medical tabular prediction,” in The Thirteenth International Conference on Learning Representations, 2025

work page 2025
[17]

DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning

D. Guo, D. Yang, H. Zhang, J. Song, R. Zhang, R. Xu, Q. Zhu, S. Ma, P. Wang, X. Bi,et al., “Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning,”arXiv preprint arXiv:2501.12948, 2025

work page internal anchor Pith review Pith/arXiv arXiv 2025
[18]

W. H. Organization,Cancer pain relief: with a guide to opioid avail- ability. World Health Organization, 1996

work page 1996
[19]

Use of opioid anal- gesics in the treatment of cancer pain: evidence-based recommendations from the eapc,

A. Caraceni, G. Hanks, S. Kaasa, M. I. Bennett, C. Brunelli, N. Cherny, O. Dale, F. De Conno, M. Fallon, M. Hanna,et al., “Use of opioid anal- gesics in the treatment of cancer pain: evidence-based recommendations from the eapc,”The lancet oncology, vol. 13, no. 2, pp. e58–e68, 2012

work page 2012
[20]

Adult cancer pain, version 3.2019, nccn clinical practice guidelines in oncology,

R. A. Swarm, J. A. Paice, D. L. Anghelescu, M. Are, J. Y . Bruce, S. Buga, M. Chwistek, C. Cleeland, D. Craig, E. Gafford,et al., “Adult cancer pain, version 3.2019, nccn clinical practice guidelines in oncology,”Journal of the National Comprehensive Cancer Network, vol. 17, no. 8, pp. 977–1007, 2019

work page 2019

[1] [1]

Pain management in lung cancer,

F. Nurwidya, E. Syahruddin, and F. Yunus, “Pain management in lung cancer,”Advances in Respiratory Medicine, vol. 84, no. 6, pp. 331–336

work page

[2] [2]

Global burden of lung cancer: Prognosis, symptom management, and challenges with pulmonary fibrosis,

D. A. Mohan Jha, D. A. Kumar, D. J. Abraham, and Huma Firdaus, “Global burden of lung cancer: Prognosis, symptom management, and challenges with pulmonary fibrosis,”International Journal of Trends in OncoScience

work page

[3] [3]

Clinical management of pain in advanced lung cancer,

C. P. Simmons, N. Macleod, and B. J. Laird, “Clinical management of pain in advanced lung cancer,”Clinical Medicine Insights: Oncology, vol. 6, p. CMO.S8360

work page

[4] [4]

Cancer pain – adult

MD Anderson Cancer Center, “Cancer pain – adult.”

work page

[5] [5]

Pain experienced by patients with ter- minal head and neck carcinoma,

Y . P. Talmi, A. Waller, M. Bercovici, Z. Horowitz, M. R. Pfeffer, A. Adunski, and J. Kronenberg, “Pain experienced by patients with ter- minal head and neck carcinoma,”Cancer: Interdisciplinary International Journal of the American Cancer Society, vol. 80, no. 6, pp. 1117–1123

work page

[6] [6]

Opioid rotation for cancer pain: rationale and clinical aspects,

S. Mercadante, “Opioid rotation for cancer pain: rationale and clinical aspects,”Cancer: Interdisciplinary International Journal of the Ameri- can Cancer Society, vol. 86, no. 9, pp. 1856–1866

work page

[7] [7]

Prediction of healthcare utilization following an episode of physical therapy for musculoskeletal pain,

T. A. Lentz, J. M. Beneciuk, and S. Z. George, “Prediction of healthcare utilization following an episode of physical therapy for musculoskeletal pain,”BMC Health Services Research, vol. 18, no. 1, p. 648

work page

[8] [8]

Cluster-then-classify methodology for the identification of pain episodes in chronic diseases,

J. Galvez-Goicurla, J. Pagan, A. B. Gago-Veiga, J. M. Moya, and J. L. Ayala, “Cluster-then-classify methodology for the identification of pain episodes in chronic diseases,”IEEE Journal of Biomedical and Health Informatics, vol. 26, no. 5, pp. 2339–2350

work page

[9] [9]

Predictive modeling for end-of-life pain outcome using electronic health records,

M. K. Lodhi, J. Stifter, Y . Yao, R. Ansari, G. M. Keenan, D. J. Wilkie, and A. A. Khokhar, “Predictive modeling for end-of-life pain outcome using electronic health records,” inAdvances in Data Mining: Appli- cations and Theoretical Aspects: 15th Industrial Conference, ICDM 2015, Hamburg, Germany, July 11-24, 2015, Proceedings 15, pp. 56–68, Springer, 2015

work page 2015

[10] [10]

Predicting the risk of cancer in adults using supervised machine learning: a scoping review,

A. A. Alfayez, H. Kunz, and A. G. Lai, “Predicting the risk of cancer in adults using supervised machine learning: a scoping review,”BMJ open, vol. 11, no. 9, p. e047755

work page

[11] [11]

Machine learning approaches to predict symptoms in people with cancer: Systematic review,

N. Zeinali, N. Youn, A. Albashayreh, W. Fan, and S. Gilbertson White, “Machine learning approaches to predict symptoms in people with cancer: Systematic review,”JMIR cancer, vol. 10, p. e52322

work page

[12] [12]

Clinical relevance of deep learning models in predicting the onset timing of cancer pain exacerbation,

Y . H. Bang, Y . H. Choi, M. Park, S.-Y . Shin, and S. J. Kim, “Clinical relevance of deep learning models in predicting the onset timing of cancer pain exacerbation,”Scientific Reports, vol. 13, no. 1, p. 11501

work page

[13] [13]

Predic- tion of cancer symptom trajectory using longitudinal electronic health record data and long short-term memory neural network,

S. Chae, W. N. Street, N. Ramaraju, and S. Gilbertson-White, “Predic- tion of cancer symptom trajectory using longitudinal electronic health record data and long short-term memory neural network,”JCO clinical cancer informatics, vol. 8, p. e2300039

work page

[14] [14]

Large language models in medicine,

A. J. Thirunavukarasu, D. S. J. Ting, K. Elangovan, L. Gutierrez, T. F. Tan, and D. S. W. Ting, “Large language models in medicine,”Nature medicine, vol. 29, no. 8, pp. 1930–1940, 2023

work page 1930

[15] [15]

Table meets llm: Can large language models understand structured table data? a benchmark and empirical study,

Y . Sui, M. Zhou, M. Zhou, S. Han, and D. Zhang, “Table meets llm: Can large language models understand structured table data? a benchmark and empirical study,” inProceedings of the 17th ACM International Conference on Web Search and Data Mining, pp. 645–654, 2024

work page 2024

[16] [16]

Small models are llm knowledge triggers for medical tabular prediction,

J. Yan, J. Chen, C. Hu, B. Zheng, Y . Hu, J. Sun, and J. Wu, “Small models are llm knowledge triggers for medical tabular prediction,” in The Thirteenth International Conference on Learning Representations, 2025

work page 2025

[17] [17]

DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning

D. Guo, D. Yang, H. Zhang, J. Song, R. Zhang, R. Xu, Q. Zhu, S. Ma, P. Wang, X. Bi,et al., “Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning,”arXiv preprint arXiv:2501.12948, 2025

work page internal anchor Pith review Pith/arXiv arXiv 2025

[18] [18]

W. H. Organization,Cancer pain relief: with a guide to opioid avail- ability. World Health Organization, 1996

work page 1996

[19] [19]

Use of opioid anal- gesics in the treatment of cancer pain: evidence-based recommendations from the eapc,

A. Caraceni, G. Hanks, S. Kaasa, M. I. Bennett, C. Brunelli, N. Cherny, O. Dale, F. De Conno, M. Fallon, M. Hanna,et al., “Use of opioid anal- gesics in the treatment of cancer pain: evidence-based recommendations from the eapc,”The lancet oncology, vol. 13, no. 2, pp. e58–e68, 2012

work page 2012

[20] [20]

Adult cancer pain, version 3.2019, nccn clinical practice guidelines in oncology,

R. A. Swarm, J. A. Paice, D. L. Anghelescu, M. Are, J. Y . Bruce, S. Buga, M. Chwistek, C. Cleeland, D. Craig, E. Gafford,et al., “Adult cancer pain, version 3.2019, nccn clinical practice guidelines in oncology,”Journal of the National Comprehensive Cancer Network, vol. 17, no. 8, pp. 977–1007, 2019

work page 2019