Breakthrough Cancer Pain Clinical Features and Differential Opioids Response: A Machine Learning Approach in Patients With Cancer From the IOPS-MS Study

Francesco Pantano; Paolo Manca; Grazia Armento; Tea Zeppola; Angelo Onorato; Michele Iuliani; Sonia Simonetti; Bruno Vincenzi; Daniele Santini; Sebastiano Mercadante; Paolo Marchetti; Arturo Cuomo; Augusto Caraceni; Rocco Domenico Mediati; Renato Vellucci; Massimo Mammucari; Silvia Natoli; Marzia Lazzari; Mario Dauri; Claudio Adile; Mario Airoldi; Giuseppe Azzarello; Livio Blasi; Bruno Chiurazzi; Daniela Degiovanni; Flavio Fusco; Vittorio Guardamagna; Simeone Liguori; Loredana Palermo; Sergio Mameli; Francesco Masedu; Teresita Mazzei; Rita Maria Melotti; Valentino Menardo; Danilo Miotti; Stefano Moroso; Gaetano Pascoletti; Stefano De Santis; Remo Orsetti; Alfonso Papa; Sergio Ricci; Elvira Scelzi; Michele Sofia; Federica Aielli; Alessandro Valle; Giuseppe Tonini

doi:10.1200/PO.20.00158

. 2020 Nov 4;4:PO.20.00158. doi: 10.1200/PO.20.00158

Breakthrough Cancer Pain Clinical Features and Differential Opioids Response: A Machine Learning Approach in Patients With Cancer From the IOPS-MS Study

Francesco Pantano ¹, Paolo Manca ^1,², Grazia Armento ^1,^✉, Tea Zeppola ¹, Angelo Onorato ¹, Michele Iuliani ¹, Sonia Simonetti ¹, Bruno Vincenzi ¹, Daniele Santini ¹, Sebastiano Mercadante ³, Paolo Marchetti ⁴, Arturo Cuomo ⁵, Augusto Caraceni ⁶, Rocco Domenico Mediati ⁷, Renato Vellucci ⁷, Massimo Mammucari ⁸, Silvia Natoli ⁹, Marzia Lazzari ⁹, Mario Dauri ⁹, Claudio Adile ³, Mario Airoldi ¹⁰, Giuseppe Azzarello ¹¹, Livio Blasi ¹², Bruno Chiurazzi ¹³, Daniela Degiovanni ¹⁴, Flavio Fusco ¹⁵, Vittorio Guardamagna ¹⁶, Simeone Liguori ¹⁷, Loredana Palermo ¹⁸, Sergio Mameli ¹⁹, Francesco Masedu ²⁰, Teresita Mazzei ²¹, Rita Maria Melotti ²², Valentino Menardo ²³, Danilo Miotti ²⁴, Stefano Moroso ²⁵, Gaetano Pascoletti ²⁶, Stefano De Santis ²⁷, Remo Orsetti ²⁸, Alfonso Papa ²⁹, Sergio Ricci ³⁰, Elvira Scelzi ³¹, Michele Sofia ³², Federica Aielli ³³, Alessandro Valle ³⁴, Giuseppe Tonini ¹

^¹Medical Oncology Department, Campus Bio-Medico University of Rome, Rome, Italy

^²IRCCS Istituto Nazionale Tumori, Milan, Italy

^³Anesthesia and Intensive Care and Pain Relief and Supportive Care, La Maddalena, Palermo, Italy

^⁴Molecular and Clinical Medicine Medical Oncology, La Sapienza University of Rome, Rome, Italy

^⁵Anesthesiology, Resuscitation, and Pain Therapy Department, National Cancer Institute, IRCCS Foundation Pascale, Naples, Italy

^⁶Palliative Care, Pain Therapy, and Rehabilitation, National Cancer Institute, IRCCS Foundation, Milan, Italy

^⁷Palliative Care and Pain Therapy Unit, Careggi Hospital, Florence, Italy

^⁸Primary Care Unit, ASL RM1, Rome, Italy

^⁹Department of Clinical Science and Translational Medicine, University of Rome Tor Vergata, Rome, Italy

^¹⁰Second Medical Oncology Division, Città della Salute e della Scienza Hospital of Turin, Turin, Italy

^¹¹Medical Specialties Department, Oncology and Oncologic Hematology, ASL 13 Mirano, Venice, Italy

^¹²Medical Oncology Unit, ARNAS Ospedale Civico Di Cristina Benfratelli, Palermo, Italy

^¹³Medical Oncology, AORN Cardarelli, Naples, Italy

^¹⁴Palliative Care Unit, ASL AL, Casale Monferrato, Italy

^¹⁵Palliative Care Unit, Department of Primary and Community Care, ASL 3 Genovese, Genoa, Italy

^¹⁶Palliative Care and Pain Therapy Unit, European Oncology Institute IRCCS, Milan, Italy

^¹⁷Palliative Care and Pain Therapy Unit, Papa Giovanni XXIII Hospital, Bergamo, Italy

^¹⁸Medical Oncology Unit, National Cancer Research Center “Giovanni Paolo II”, Bari, Italy

^¹⁹Pain Therapy Unit, “A. Businco” Hospital, ASL 8, Cagliari, Italy

^²⁰Department of Biotechnological and Applied Clinical Sciences, Section of Clinical Epidemiology and Environmental Medicine, University of L'Aquila, L'Aquila, Italy

^²¹Section of Clinical Pharmacology and Oncology, Department of Health Sciences, University of Florence, Florence, Italy

^²²Department of Medicine and Surgery Sciences, University of Bologna, Bologna, Italy

^²³PainTherapy, S. Croce e Carle Hospital, Cuneo, Italy

^²⁴Pain Therapy ICS Maugeri, IRCCS Foundation, Pavia, Italy

^²⁵Medical Oncology, Azienda Sanitaria Universitaria Integrata di Trieste, Trieste, Italy

^²⁶Medical Oncology, Azienda Sanitaria Universitaria Integrata di Udine, Udine, Italy

^²⁷Palliative Care and Oncologic Pain Service, S. Camillo-Forlanini Hospital, Rome, Italy

^²⁸Pain Medicine Unit, S. Camillo-Forlanini Hospital, Rome, Italy

^²⁹Pain Relief, A.O. Dei Colli, Monaldi Hospital, Naples, Italy

^³⁰Division of Medical Oncology, Department of Oncology, S. Chiara University Hospital, Pisa, Italy

^³¹Medical Oncology, Castelfranco Veneto Hospital, Treviso, Italy

^³²Department of Palliative Care, Hospice and Pain Therapy Unit, “G. Salvini” Hospital, Milan, Italy

^³³Department of Biotechnological and Applied Clinical Sciences, University of L'Aquila, L'Aquila, Italy

^³⁴Palliative Care, FARO Foundation, Turin, Italy

^✉

Grazia Armento, MD, Medical Oncology Department, Campus Bio-Medico University of Rome, Via Alvaro Del Portillo 21, Rome 128, Italy; e-mail: g.armento@unicampus.it.

PMCID: PMC7713587 PMID: 33283139

Abstract

PURPOSE

A large proportion of patients with cancer suffer from breakthrough cancer pain (BTcP). Several unmet clinical needs concerning BTcP treatment, such as optimal opioid dosages, are being investigated. In this analysis the hypothesis, we explore with an unsupervised learning algorithm whether distinct subtypes of BTcP exist and whether they can provide new insights into clinical practice.

METHODS

Partitioning around a k-medoids algorithm on a large data set of patients with BTcP, previously collected by the Italian Oncologic Pain Survey group, was used to identify possible subgroups of BTcP. Resulting clusters were analyzed in terms of BTcP therapy satisfaction, clinical features, and use of basal pain and rapid-onset opioids. Opioid dosages were converted to a unique scale and the BTcP opioids-to-basal pain opioids ratio was calculated for each patient. We used polynomial logistic regression to catch nonlinear relationships between therapy satisfaction and opioid use.

RESULTS

Our algorithm identified 12 distinct BTcP clusters. Optimal BTcP opioids-to-basal pain opioids ratios differed across the clusters, ranging from 15% to 50%. The majority of clusters were linked to a peculiar association of certain drugs with therapy satisfaction or dissatisfaction. A free online tool was created for new patients’ cluster computation to validate these clusters in future studies and provide handy indications for personalized BTcP therapy.

CONCLUSION

This work proposes a classification for BTcP and identifies subgroups of patients with unique efficacy of different pain medications. This work supports the theory that the optimal dose of BTcP opioids depends on the dose of basal opioids and identifies novel values that are possibly useful for future trials. These results will allow us to target BTcP therapy on the basis of patient characteristics and to define a precision medicine strategy also for supportive care.

INTRODUCTION

Breakthrough cancer pain (BTcP) is a common event that affects a considerable proportion of patients with cancer.¹ A variety of definitions for BTcP have been proposed^2,3. According to the Italian Oncologic Pain Survey (IOPS) study group,⁴ BTcP should be defined “as a relevant change in pain intensity of severe intensity in patients who receive an effective treatment with opioids.”⁴^(p963) Nevertheless, despite this unique definition, BTcP encloses a wide range of manifestations that differ, among other features, in intensity, duration, frequency, and triggering events. BTcP represents a clinically relevant condition with a negative impact on the patient’s quality of life. In the majority, it is difficult to achieve an acceptable degree of relief because patients with cancer have complex pain syndromes. These patients often require more intense therapeutic protocols and, therefore, more time may be required to achieve adequate pain control.⁵

At present, several gaps exist in the knowledge of BTcP management. These partially unanswered questions, among others, concern the optimal drug administration route, pharmacokinetics, the balance between rapid-onset and slow-onset opioids, and the eventual difference of BTcP response deriving from clinical features, such as stage, primary site, or metastases.

In this analysis, it was thus hypothesized that the unique BTcP definition might actually enclose diverse pathologic entities, possibly with peculiar clinical needs and specific responses to drugs. To explore this supposition, we used novel multiparametric artificial intelligence algorithms that can simultaneously analyze different clinical features and identify the existence of shared patterns. These so-called unsupervised learning algorithms have already been extensively used, for example, for the identification of breast cancer subtypes.⁶ Nevertheless, to our knowledge, no authors have yet tried to explore the issue of BTcP management using these techniques.

To fulfill this purpose, we used data that were collected by the IOPS group in a large, multicentric national study^5,7,8 that enrolled 4,056 patients from 32 centers with opioid-controlled basal pain suffering from BTcP. This work is therefore a secondary analysis of the IOPS group survey that aims to identify novel subtypes of BTcP through the use of unsupervised learning algorithms.

METHODS

Patients’ Enrollment and Data Collection

Details concerning the enrollment of patients are extensively described in the main paper from the IOPS group.⁵ In brief, local ethical committees approved the protocol, and written informed consent was obtained from each patient. Interviews were performed in different settings, in particular, oncology, pain therapy, palliative care, and radiotherapy. Patients were ≥ 18 years of age; diagnosed with cancer at any stage; had stable background pain in the last week with an intensity of, at most, 4 on a numerical scale from 0 to 10; and had episodes of BTcP with an intensity ≥ 5, clearly distinguished from background pain. The definition of BTP was a transitory pain exacerbation of moderate to severe intensity that occurs spontaneously or is predictably well distinguished from background pain. Exclusion criteria were the absence of a cancer diagnosis, uncontrolled background pain (> 4 on a numerical scale of 0 to 10), or no relevant increases in pain intensity (< 5) that could be interpreted as BTcP episodes. Patients who were unable to provide information about the data required for the study as a result of either cognitive failure or terminal disease were also excluded. A comprehensive list of clinical variables was collected for each patient that consisted of basal pain and BTcP site, duration, frequency, intensity, relieving factors, triggers, drugs, primary cancer site and stage, and concomitant symptoms for a total of 1,086 variables. Interviews were registered by collecting personnel in a closed online form and centrally stored.

Therapy Satisfaction

The association of each clinical feature with satisfaction toward BTcP therapy was investigated using a simple logistic regression. Therapy satisfaction was expressed as a binomial outcome. We used the false discovery rate method⁹ to correct P values for multiple comparisons. Features with less than 5% of missing data and associated with a corrected P value of < .1 and, for categorical features, a log₂ odds ratio greater than 1 were used to build a multivariable logistic regression. To investigate the correlation—simultaneously for all patients and on the same scale—between the amount of opioids used and BTcP therapy satisfaction, all doses of opioid drugs were converted¹⁰ to equivalent intravenous morphine doses and expressed as a total daily dose: one for BTcP-directed opioids and for basal pain opioids. Doses were converted to intravenous morphine doses and not to oral morphine doses because intravenous morphine has been increasingly used in different clinical settings and would provide more interpretable graphics in Results. Moreover, to explore the interaction of fast-acting and long-acting opioid dosages, we calculated for each patient the BTcP opioids–to–basal pain opioids ratio (OpR). A polynomial logistic regression was used to catch nonlinear relationships between opioid doses and therapy satisfaction.

Cluster Computation and Visualization

Features defining the clinical characteristics of BTcP were selected to perform clusters computation. Features with missing data that accounted for more than 5% of patients were excluded. The above-mentioned features were used to calculate a dissimilarity matrix using a cluster package¹¹; as features were also composed of non-numeric variables, we used the Gower metric¹² for dissimilarity matrix calculation. Partitioning around the k-medoids¹³ algorithm was used to compute clusters using the dissimilarity matrix as an input. The algorithm was run different times using a range of cluster numbers (2-30). Silhouette statistics¹⁴ were calculated for each run, and the optimal number of clusters was manually picked as being that with the best trade-off between silhouette statistics and reasonable clinical interpretation. We used complete-linkage hierarchical clustering with 12 clusters, along with Rand index calculation, to explore the replicability of clusters with a different algorithm.

t-Distributed stochastic neighbor embedding¹⁵ algorithm was used to project dissimilarities between patients in a bidimensional space, with closer points representing patients with more similar clinical BTcP features. An online tool allows for the repetition of the performed classification on new sets of patients.¹⁶

Clusters Analysis

t-Test, Mann-Whitney test, and χ² test were used to assess the association of parametric, nonparametric, and categorical features, respectively, with each cluster. The false discovery rate method was used to correct P values for multiple comparisons. Therapy satisfaction was investigated separately for each cluster, as previously described, for all samples.

Data Handling

Data were imported and analyzed using R (v3.5.2).¹⁷ The following packages were used for the analyses: dplyr,¹⁸ cluster,¹¹ Rtsne,¹⁹ ggplot2,²⁰ gmodels,²¹ Rmisc,²² epiR,²³ mgcv,²⁴ knitr,²⁵ RColorBrewer,²⁶ and mgcViz.²⁷

RESULTS

Patient Characteristics

A total of 4,016 patients were enrolled in the study during a period of 24 months. Men accounted for 54.8% of the total, and mean age was 64.6 years (range, 18-97 years). The majority of visits were performed for oncologic (52.0%) and pain therapy (29.5%) purposes. Together, inpatient (37.3%) and outpatient (34.3%) settings accounted for more than one half of visits. The most common cancer primary organs were the lung (24.0%), breast (11.3%), pancreas (8.3%), and colon 7.5%. Mean Karnofsky performance status was 48. According to the inclusion criteria, basal pain was generally controlled; mean basal pain numeric rating scale was 3.0. Patients’ characteristics are listed in Table 1.

TABLE 1.

Patients’ Characteristics

graphic file with name PO.20.00158t1.jpg

Open in a new tab

Cluster Computation

To investigate whether subtypes of BTcP exist, we used BTcP features to build an unsupervised clustering model. The number of BTcP episodes, BTcP peaks duration, BTcP type, BTcP intensity, number of days since the beginning of BTcP episodes, the eventual benefit from pharmacotherapy, the eventual benefit from rest, and whether BTcP was enhanced by movements were the eight BTcP-defining variables selected for the final model, which was built with the k-medoids algorithm. We chose 12 as an optimal trade-off between the average width of clusters silhouette (0.45; Appendix Fig A1A) and the interpretability of the clusters themselves. The average internal dissimilarity was acceptably low, ranging between 0.05 and 0.16. A Rand index of 0.89 was obtained comparing results from PAM (partition around medoids) clustering and hierarchical clustering, which suggests good stability of the partitioning with different methods (Appendix Fig A1B).

Figure 1 shows the algorithm used to define BTP (Fig 1A) and a 2-dimensional t-distributed stochastic neighbor embedding projection of all patients, colored by their clusters (Fig 1B). An online tool is available for the classification of new patients according to our method.¹⁶

Characteristics of BTcP Clusters

We analyzed the enrichment of the eight BTcP-defining variables and of other clinical features among the clusters. A description of each cluster is available in Table 2. A summary of BTcP features according to cluster is presented in Figure 2.

TABLE 2.

Description of Clusters

Open in a new tab

FIG 2. — Defining features of the 12 breakthrough cancer pain (BTcP) clusters. Plots represent in order: (A) BTcP intensity using numeric rating scale, (B) BTcP peak duration, (C) BTcP type, (D) number of BTcP events per day, (E) presence of benefit in BTcP management with pharmacotherapy, (F) presence of benefit in BTcP management with rest, (G) presence of BTcP activation with movements, and (H) days since BTcP episodes started.

BTcP Therapy Satisfaction

Finally, we tried to assess what influenced patient-reported BTcP therapy satisfaction. After converting the opioid dose to a unique scale—corresponding to the equivalent dose of intravenous morphine—we investigated using a nonlinear model the influence of basal opioid dose, BTcP opioid dose, and OpR on patient-reported therapy satisfaction. Of interest, whereas the basal opioid dose did not demonstrate any great impact on therapy satisfaction (Fig 3C), and BTcP opioid dose showed some irregular peaks of satisfaction with CIs that often reached the indifference line (Fig 3B), OpR seemed to depict a clear, optimal peak between 0.4 and 0.45 (Fig 3A). This roughly corresponds to a daily dose of sublingual fentanyl 100 μg for BTcP and a daily dose of oral morphine 30 mg as the basal opioid dose.

We separately performed the same analyses on previously defined clusters (Fig 3D). Of interest, not all the clusters showed the same relationship between OpR and satisfaction. For clusters 1, 6, and 10, the satisfaction seemed to grow indefinitely with an increase of the OpR opioids, whereas clusters 7, 8, and 11 seemed to have clear, optimal peaks of OpR. Despite the interpretation being challenged by some large CIs, we can say from these data that, depending on the cluster, optimal OpR ranges from 15% to 50%.

DISCUSSION

This paper demonstrates a novel approach for the investigation of BTcP. Our findings identified 12 subtypes of BTcP with peculiar response to drugs and clinical presentation. We acknowledge that this study was not designed to perform this analysis and, moreover, that the large number of clusters might interfere with their interpretability and clinical utility. Nevertheless, this study represents a proof-of-concept for this investigational approach.

Some of our findings might already provide some indication for future clinical practice. First, it seems that an optimal ratio between BTcP opioids and basal pain opioids exists. Using our same data, another group proposed 0.20 (one fifth) as the optimal ratio.⁵ Nevertheless, they used a frequentistic approach, as 0.20 is simply the most common ratio among the cohort. Our group instead modeled the ratio toward an outcome (BTcP therapy satisfaction) and highlighted a peak of satisfaction within a ratio range of 0.40-0.45. What seems clear, though, is that such an optimal level exists. This possibly suggests that, instead of starting BTcP opioid titration with the lowest possible dosage, as proposed previously,²⁸ titration could start immediately with an optimal opioid dosage. Moreover, cluster analysis reveals that this ratio might not be the same for all patients: Some patients might benefit from a higher BTcP opioid dosage (cluster 2 and 7), whereas others might benefit from a lower one (cluster 11). Finally, for some patients, we did not observe an upper threshold for this ratio (cluster 1, 6, and 9), perhaps pointing out patients for whom a strong BTcP opioid dosage increase is required.

The interpretation is made difficult by the multiple associations and, at this stage, is not mature enough to suggest any immediate change in clinical practice. However, we made available an online free tool¹⁶ that allows for the classification of new patients according to our algorithm and returns a proposed BTcP therapy which depends on the patient cluster optimal OpR and basal opioid dose. We suggest that this tool might be used in the future to prospectively validate the clinical importance of our clusters in clinical practice and to compare our proposed opioid dosages in settings different from ours.

The presence of distinct BTcP phenotypes, each one associated with specific clinical features, could also be a reflection of diverse underlying pathophysiologic mechanisms: Our work suggests that preclinical research might gain insight into these possible differences and help the development of a tailored therapy also for BTcP.

The main limitation of our study is the appropriateness of collected data for the scope of our work. We believe that a prospective study specifically designed for the investigation of BTcP clusters—possibly with long-term follow-up and therapy success outcomes and not limited to a single timepoint evaluation—might enable a clearer identification of distinct clusters. Moreover, our approach lacks an external validation of cluster consistencies and reproducibility. Nevertheless, these limitations do not interfere with the main scope of our paper, which was to offer a proof of concept for an innovative approach for BTcP management.

In conclusion, this work identifies criteria for optimal BTcP opioids therapy personalization and offers a reproducible classification for the enrollment and stratification of patients in future BTcP trials.

Appendix

FIG A1. — (A) Silhouette statistics for clusters 2 to 30. The appropriate number of clusters for additional analyses was found to be 12, an optimal trade-off between the average width of clusters silhouette (0.45) and the interpretability of clusters themselves. (B) Heatmap showing the concordance of PAM clustering with complete-linkage hierarchical clustering. The Rand index, which compares the replicability of the two algorithms, was 0.89.

EQUAL CONTRIBUTION

F.P. and P.M. contributed equally to this work.

PRIOR PRESENTATION

Presented at the ASCO 2020 Virtual Annual Meeting, May 29-31, 2020.

AUTHOR CONTRIBUTIONS

Conception and design: Francesco Pantano, Paolo Manca, Grazia Armento, Tea Zeppola, Bruno Vincenzi, Daniele Santini, Sebastiano Mercadante, Paolo Marchetti, Augusto Caraceni, Rocco Domenico Mediati, Renato Vellucci, Massimo Mammucari, Mario Airoldi, Giuseppe Azzarello, Bruno Chiurazzi, Simeone Liguori, Loredana Palermo, Rita Maria Melotti, Michele Sofia, Giuseppe Tonini

Financial support: Loredana Palermo, Giuseppe Tonini

Administrative support: Loredana Palermo, Giuseppe Tonini

Provision of study materials or patients: Bruno Vincenzi, Arturo Cuomo, Augusto Caraceni, Rocco Domenico Mediati, Silvia Natoli, Marzia Lazzari, Mario Dauri, Livio Blasi, Bruno Chiurazzi, Loredana Palermo, Sergio Mameli, Danilo Miotti, Stefano Moroso, Gaetano Pascoletti, Stefano De Santis, Remo Orsetti, Sergio Ricci, Federica Aielli, Giuseppe Tonini

Collection and assembly of data: Grazia Armento, Tea Zeppola, Angelo Onorato, Sonia Simonetti, Paolo Marchetti, Arturo Cuomo, Augusto Caraceni, Rocco Domenico Mediati, Silvia Natoli, Marzia Lazzari, Mario Dauri, Claudio Adile, Mario Airoldi, Giuseppe Azzarello, Livio Blasi, Bruno Chiurazzi, Daniela Degiovanni, Flavio Fusco, Vittorio Guardamagna, Simeone Liguori, Sergio Mameli, Teresita Mazzei, Valentino Menardo, Danilo Miotti, Stefano Moroso, Gaetano Pascoletti, Stefano De Santis, Alfonso Papa, Sergio Ricci, Elvira Scelzi, Michele Sofia, Federica Aielli, Alessandro Valle, Giuseppe Tonini

Data analysis and interpretation: Francesco Pantano, Paolo Manca, Grazia Armento, Tea Zeppola, Michele Iuliani, Bruno Vincenzi, Paolo Marchetti, Arturo Cuomo, Augusto Caraceni, Claudio Adile, Mario Airoldi, Simeone Liguori, Francesco Masedu, Stefano De Santis, Remo Orsetti, Michele Sofia, Giuseppe Tonini

Manuscript writing: All authors

Final approval of manuscript: Conception and design

Accountable for all aspects of the work: Conception and design

AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST

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/po/author-center.

Open Payments is a public database containing information reported by companies about payments made to US-licensed physicians (Open Payments).

Paolo Manca

Patents, Royalties, Other Intellectual Property: Patent for a method for the identification of gene panels optimal for TMB estimation (Inst)

Grazia Armento

Consulting or Advisory Role: Novartis

Travel, Accommodations, Expenses: Teva

Bruno Vincenzi

Speakers' Bureau: PharmaMar

Paolo Marchetti

Honoraria: AstraZeneca, Boehringer Ingelheim, Bristol Myers Squibb, Celgene, Eisai, Incyte, Merck Sharp & Dohme, Novartis, Pierre Fabre, Pfizer, Roche

Consulting or Advisory Role: AstraZeneca, Boehringer Ingelheim, Bristol Myers Squibb, Merck Sharp & Dohme, Novartis, Pierre Fabre, Roche

Research Funding: AstraZeneca (Inst), Bristol Myers Squibb (Inst), Incyte (Inst), Eli Lilly (Inst), Merck Serono (Inst), Merck Sharp & Dohme (Inst), Novartis (Inst), NanoString Technologies (Inst), Pierre Fabre (Inst), Pfizer (Inst), Roche (Inst), Takeda (Inst)

Travel, Accommodations, Expenses: Boehringer Ingelheim, Bristol Myers Squibb, Incyte, Merck Sharp & Dohme, Roche

Augusto Caraceni

Consulting or Advisory Role: Angelini Pharma, Shionogi Molteni, Kyowa Hakko Kirin

Rocco Domenico Mediati

Speakers' Bureau: Angelini Pharma, Kyowa Hakko Kirin

Renato Vellucci

Travel, Accommodations, Expenses: Grünenthal Group

Silvia Natoli

Consulting or Advisory Role: Grünenthal Italia, Sandoz, Angelini Pharma

Speakers' Bureau: Mylan

Travel, Accommodations, Expenses: AMS Group, Sandoz

Giuseppe Azzarello

Consulting or Advisory Role: Merck

Vittorio Guardamagna

Consulting or Advisory Role: Istituto Gentili

Stefano Moroso

Consulting or Advisory Role: Novartis, Janssen-Cilag

Giuseppe Tonini

Consulting or Advisory Role: Novartis, Molteni Farmaceutici, Roche, Pierre Fabre, Italfarmaco

Research Funding: PharmaMar (Inst), Novartis (Inst)

No other potential conflicts of interest were reported.

REFERENCES

1.Caraceni A, Martini C, Zecca E, et al. Breakthrough pain characteristics and syndromes in patients with cancer pain. An international survey. Palliat Med. 2004;18:177–183. doi: 10.1191/0269216304pm890oa. [DOI] [PubMed] [Google Scholar]
2.Portenoy RK, Hagen NA. Breakthrough pain: Definition, prevalence and characteristics. Pain. 1990;41:273–281. doi: 10.1016/0304-3959(90)90004-W. [DOI] [PubMed] [Google Scholar]
3.Portenoy RK, Payne D, Jacobsen P. Breakthrough pain: Characteristics and impact in patients with cancer pain. Pain. 1999;81:129–134. doi: 10.1016/s0304-3959(99)00006-8. [DOI] [PubMed] [Google Scholar]
4.Mercadante S, Marchetti P, Cuomo A, et al. Breakthrough pain and its treatment: Critical review and recommendations of IOPS (Italian Oncologic Pain Survey) expert group. Support Care Cancer. 2016;24:961–968. doi: 10.1007/s00520-015-2951-y. [Erratum: Support Care Cancer 25:2673-2674] [DOI] [PubMed] [Google Scholar]
5.Mercadante S, Marchetti P, Cuomo A, et al. Factors influencing the clinical presentation of breakthrough pain in cancer patients. Cancers (Basel) 2018;10:175. doi: 10.3390/cancers10060175. [DOI] [PMC free article] [PubMed] [Google Scholar]
6.Sørlie T, Perou CM, Tibshirani R, et al. Gene expression patterns of breast carcinomas distinguish tumor subclasses with clinical implications. Proc Natl Acad Sci USA. 2001;98:10869–10874. doi: 10.1073/pnas.191367098. [DOI] [PMC free article] [PubMed] [Google Scholar]
7.Mercadante S, Marchetti P, Cuomo A, et al. Breakthrough cancer pain: Preliminary data of the Italian Oncologic Pain Multisetting Multicentric Survey (IOPS-MS) Adv Ther. 2017;34:120–135. doi: 10.1007/s12325-016-0440-4. [DOI] [PMC free article] [PubMed] [Google Scholar]
8.Mercadante S, Lazzari M, Reale C, et al. Italian Oncological Pain Survey (IOPS): A multicentre Italian study of breakthrough pain performed in different settings. Clin J Pain. 2015;31:214–221. doi: 10.1097/AJP.0000000000000161. [DOI] [PubMed] [Google Scholar]
9.Benjamini Y, Hochberg Y. Controlling the false discovery rate: A practical and powerful approach to multiple testing. J R Stat Soc Ser B (Method) 1995;57:289–300. [Google Scholar]
10.National Comprehensive Cancer Network Clinical practice guidelines: Adult cancer pain. https://www.nccn.org/professionals/physician_gls/pdf/pain.pdf
11.Maechler M, Rousseeuw P, Struyf A, et al. cluster: Cluster analysis basics and extensions. R package version 2.1.0. https://cran.r-project.org/web/packages/cluster/cluster.pdf
12.Gower JC. A general coefficient of similarity and some of its properties. Biometrics. 1971;27:857–871. [Google Scholar]
13.Dodge Y.(ed)Clustering by means of MedoidsinStatistical Data Analysis Based on the L₁-Norm and Related Methods North-Holland; Amsterdam, Springer; 405–416.1987 [Google Scholar]
14.Rousseeuw PJ. Silhouettes: A graphical aid to the interpretation and validation of cluster analysis. J Comput Appl Math. 1987;20:53–65. [Google Scholar]
15.van der Maaten LJP, Hinton GE. Visualizing high-dimensional data using t-SNE. J Mach Learn Res. 2008;9:2579–2605. [Google Scholar]
16. UCBM BTcP Clusters. https://mancapaolo.shinyapps.io/UCBM_BTcPclusters.
17.R Core Team . R: A language and environment for statistical computing. Vienna, Austria: R Foundation for Statistical Computing; https://www.R-project.org/ [Google Scholar]
18.Hadley W, Romain F, Lionel H, et al. dplyr: A grammar of data manipulation. R package version 0.8.0.1. https://cran.r-project.org/web/packages/dplyr/index.html
19.Krijthe JH. Rtsne: T-distributed stochastic neighbor embedding using a Barnes-Hut implementation. https://github.com/jkrijthe/Rtsne
20.Wickham H. ggplot2 Elegant Graphics for Data Analysis. ed 2. New York, NY: Springer-Verlag; 2016. [Google Scholar]
21.Warnes GR, Bolker B, Lumley T, et al. gmodels: Various R programming tools for model fitting. R package version 2.18.1. https://cran.r-project.org/web/packages/gmodels/gmodels.pdf
22.Ryan M. Rmisc: Ryan miscellaneous. R package version 1.5. https://CRAN.R-project.org/package=Rmisc
23.Stevenson M., Nunes T, Heuer C, et al. epiR: Tools for the analysis of epidemiological data. R package version 0.9-99. https://cran.r-project.org/web/packages/epiR/index.html
24.Wood SN. Stable and efficient multiple smoothing parameter estimation for generalized additive models. J Am Stat Assoc. 2004;99:673–686. [Google Scholar]
25.Xie Y, Vogt A, Andrew A, et al. knitr: A general-purpose package for dynamic report generation in R. R package version 1.21. https://cran.r-project.org/web/packages/knitr/index.html
26.Neuwirth E. RColorBrewer: ColorBrewer palettes. R package version 1.1-2. https://CRAN.R-project.org/package=RColorBrewer
27.Fasiolo M, Nedellec R, Goude Y. Scalable visualisation methods for modern generalized additive models. 2018 https://arxiv.org/abs/1809.10632 arXiv:1809.10632.
28.Davies AN, Dickman A, Reid C, et al. The management of cancer-related breakthrough pain: Recommendations of a task group of the Science Committee of the Association for Palliative Medicine of Great Britain and Ireland. Eur J Pain. 2009;13:331–338. doi: 10.1016/j.ejpain.2008.06.014. [DOI] [PubMed] [Google Scholar]

[B1] 1.Caraceni A, Martini C, Zecca E, et al. Breakthrough pain characteristics and syndromes in patients with cancer pain. An international survey. Palliat Med. 2004;18:177–183. doi: 10.1191/0269216304pm890oa. [DOI] [PubMed] [Google Scholar]

[B2] 2.Portenoy RK, Hagen NA. Breakthrough pain: Definition, prevalence and characteristics. Pain. 1990;41:273–281. doi: 10.1016/0304-3959(90)90004-W. [DOI] [PubMed] [Google Scholar]

[B3] 3.Portenoy RK, Payne D, Jacobsen P. Breakthrough pain: Characteristics and impact in patients with cancer pain. Pain. 1999;81:129–134. doi: 10.1016/s0304-3959(99)00006-8. [DOI] [PubMed] [Google Scholar]

[B4] 4.Mercadante S, Marchetti P, Cuomo A, et al. Breakthrough pain and its treatment: Critical review and recommendations of IOPS (Italian Oncologic Pain Survey) expert group. Support Care Cancer. 2016;24:961–968. doi: 10.1007/s00520-015-2951-y. [Erratum: Support Care Cancer 25:2673-2674] [DOI] [PubMed] [Google Scholar]

[B5] 5.Mercadante S, Marchetti P, Cuomo A, et al. Factors influencing the clinical presentation of breakthrough pain in cancer patients. Cancers (Basel) 2018;10:175. doi: 10.3390/cancers10060175. [DOI] [PMC free article] [PubMed] [Google Scholar]

[B6] 6.Sørlie T, Perou CM, Tibshirani R, et al. Gene expression patterns of breast carcinomas distinguish tumor subclasses with clinical implications. Proc Natl Acad Sci USA. 2001;98:10869–10874. doi: 10.1073/pnas.191367098. [DOI] [PMC free article] [PubMed] [Google Scholar]

[B7] 7.Mercadante S, Marchetti P, Cuomo A, et al. Breakthrough cancer pain: Preliminary data of the Italian Oncologic Pain Multisetting Multicentric Survey (IOPS-MS) Adv Ther. 2017;34:120–135. doi: 10.1007/s12325-016-0440-4. [DOI] [PMC free article] [PubMed] [Google Scholar]

[B8] 8.Mercadante S, Lazzari M, Reale C, et al. Italian Oncological Pain Survey (IOPS): A multicentre Italian study of breakthrough pain performed in different settings. Clin J Pain. 2015;31:214–221. doi: 10.1097/AJP.0000000000000161. [DOI] [PubMed] [Google Scholar]

[B9] 9.Benjamini Y, Hochberg Y. Controlling the false discovery rate: A practical and powerful approach to multiple testing. J R Stat Soc Ser B (Method) 1995;57:289–300. [Google Scholar]

[B10] 10.National Comprehensive Cancer Network Clinical practice guidelines: Adult cancer pain. https://www.nccn.org/professionals/physician_gls/pdf/pain.pdf

[B11] 11.Maechler M, Rousseeuw P, Struyf A, et al. cluster: Cluster analysis basics and extensions. R package version 2.1.0. https://cran.r-project.org/web/packages/cluster/cluster.pdf

[B12] 12.Gower JC. A general coefficient of similarity and some of its properties. Biometrics. 1971;27:857–871. [Google Scholar]

[B13] 13.Dodge Y.(ed)Clustering by means of MedoidsinStatistical Data Analysis Based on the L₁-Norm and Related Methods North-Holland; Amsterdam, Springer; 405–416.1987 [Google Scholar]

[B14] 14.Rousseeuw PJ. Silhouettes: A graphical aid to the interpretation and validation of cluster analysis. J Comput Appl Math. 1987;20:53–65. [Google Scholar]

[B15] 15.van der Maaten LJP, Hinton GE. Visualizing high-dimensional data using t-SNE. J Mach Learn Res. 2008;9:2579–2605. [Google Scholar]

[B16] 16. UCBM BTcP Clusters. https://mancapaolo.shinyapps.io/UCBM_BTcPclusters.

[B17] 17.R Core Team . R: A language and environment for statistical computing. Vienna, Austria: R Foundation for Statistical Computing; https://www.R-project.org/ [Google Scholar]

[B18] 18.Hadley W, Romain F, Lionel H, et al. dplyr: A grammar of data manipulation. R package version 0.8.0.1. https://cran.r-project.org/web/packages/dplyr/index.html

[B19] 19.Krijthe JH. Rtsne: T-distributed stochastic neighbor embedding using a Barnes-Hut implementation. https://github.com/jkrijthe/Rtsne

[B20] 20.Wickham H. ggplot2 Elegant Graphics for Data Analysis. ed 2. New York, NY: Springer-Verlag; 2016. [Google Scholar]

[B21] 21.Warnes GR, Bolker B, Lumley T, et al. gmodels: Various R programming tools for model fitting. R package version 2.18.1. https://cran.r-project.org/web/packages/gmodels/gmodels.pdf

[B22] 22.Ryan M. Rmisc: Ryan miscellaneous. R package version 1.5. https://CRAN.R-project.org/package=Rmisc

[B23] 23.Stevenson M., Nunes T, Heuer C, et al. epiR: Tools for the analysis of epidemiological data. R package version 0.9-99. https://cran.r-project.org/web/packages/epiR/index.html

[B24] 24.Wood SN. Stable and efficient multiple smoothing parameter estimation for generalized additive models. J Am Stat Assoc. 2004;99:673–686. [Google Scholar]

[B25] 25.Xie Y, Vogt A, Andrew A, et al. knitr: A general-purpose package for dynamic report generation in R. R package version 1.21. https://cran.r-project.org/web/packages/knitr/index.html

[B26] 26.Neuwirth E. RColorBrewer: ColorBrewer palettes. R package version 1.1-2. https://CRAN.R-project.org/package=RColorBrewer

[B27] 27.Fasiolo M, Nedellec R, Goude Y. Scalable visualisation methods for modern generalized additive models. 2018 https://arxiv.org/abs/1809.10632 arXiv:1809.10632.

[B28] 28.Davies AN, Dickman A, Reid C, et al. The management of cancer-related breakthrough pain: Recommendations of a task group of the Science Committee of the Association for Palliative Medicine of Great Britain and Ireland. Eur J Pain. 2009;13:331–338. doi: 10.1016/j.ejpain.2008.06.014. [DOI] [PubMed] [Google Scholar]

PERMALINK

Breakthrough Cancer Pain Clinical Features and Differential Opioids Response: A Machine Learning Approach in Patients With Cancer From the IOPS-MS Study

Francesco Pantano, MD, PhD

Paolo Manca, MD

Grazia Armento, MD

Tea Zeppola, MD

Angelo Onorato, MD

Michele Iuliani, PhD

Sonia Simonetti, PhD

Bruno Vincenzi, MD, PhD

Daniele Santini, MD

Sebastiano Mercadante, MD

Paolo Marchetti, MD

Arturo Cuomo, MD

Augusto Caraceni, MD

Rocco Domenico Mediati, MD

Renato Vellucci, MD

Massimo Mammucari, MD

Silvia Natoli, MD, PhD

Marzia Lazzari, MD

Mario Dauri, MD

Claudio Adile, MD

Mario Airoldi, MD

Giuseppe Azzarello, MD

Livio Blasi, MD

Bruno Chiurazzi, MD

Daniela Degiovanni, MD

Flavio Fusco, MD, PhD

Vittorio Guardamagna, MD

Simeone Liguori, MD

Loredana Palermo, PhD

Sergio Mameli, MD

Francesco Masedu, PhD

Teresita Mazzei, PhD

Rita Maria Melotti, MD

Valentino Menardo, MD

Danilo Miotti, PhD

Stefano Moroso, MD

Gaetano Pascoletti, MD

Stefano De Santis, MD

Remo Orsetti, MD

Alfonso Papa, MD

Sergio Ricci, MD

Elvira Scelzi, MD

Michele Sofia, MD

Federica Aielli, MD

Alessandro Valle, MD

Giuseppe Tonini, MD, PhD

Abstract

PURPOSE

METHODS

RESULTS

CONCLUSION

INTRODUCTION

METHODS

Patients’ Enrollment and Data Collection

Therapy Satisfaction

Cluster Computation and Visualization

Clusters Analysis

Data Handling

RESULTS

Patient Characteristics

TABLE 1.

Cluster Computation

FIG 1.

Characteristics of BTcP Clusters

TABLE 2.

FIG 2.

BTcP Therapy Satisfaction

FIG 3.

DISCUSSION

Appendix

FIG A1.

EQUAL CONTRIBUTION

PRIOR PRESENTATION

AUTHOR CONTRIBUTIONS

AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST

Paolo Manca

Grazia Armento

Bruno Vincenzi