3
When a Decision Must Be Made: Role of Computer Modeling in Clinical Cancer Research Rebecca A. Miksad, Beth Israel Deaconess Medical Center, Harvard Medical School, Boston, MA See accompanying article doi: 10.1200/JCO.2010.33.8020 Every day, multidisciplinary oncology teams make dozens of treat- ment decisions that may have a tremendous impact on a patient’s survival and quality of life. Made with the best of intentions, these decisions are informed by basic science and clinical research findings, clinical experience, and health policy. All too often, results from the gold standard of clinical trial research, a randomized controlled trial (RCT), that fit the specific details of the patient’s situation are not available to guide these decisions. Although this data gap occurs at times for all cancers, it is a constant limitation for less common and biologically heterogeneous diseases. For these cancers, such as those of the biliary tract, practical time and expense limitations restrict the number and combinations of therapeutic strategies evaluated, the follow-up duration, and the pop- ulations studied in RCTs. 1 And even in the most common cancers, the costly failure of multiple trials that involve thousands of patients to move cancer care forward has raised the need for alternate re- search paradigms. 2-9 Enter computer modeling as a method to bridge current knowledge gaps and to advance cancer clinical care and research. When performed correctly—rigorously developed, calibrated, and validated— computer modeling can maximize the information that is gained from current clinical, basic science, and epidemio- logic research efforts to facilitate informed clinical and health policy decisions. 10 This power stems from the ability of computer models to produce novel comparative effectiveness findings, extend trial results to longer time horizons, expand study findings to new populations, and refine expected outcomes. Last, but not least, com- puter models may also help differentiate between those scientific and clinical questions for which an RCT would be preferred but is not vital for decision making, and those questions for which the expense, time, and patient effort of an RCT is absolutely required to improve out- comes and to guide treatment and policy decisions. 11-20 In the article that accompanies this editorial, Wang et al 21 used survival model techniques to predict the benefit of adjuvant chemo- therapy and chemoradiotherapy for patients with resected gallbladder cancer. Although the prognosis for these patients is usually grim and the need for an effective treatment is great, there is a paucity of pub- lished information to guide adjuvant therapy choices. 22-25 However, despite this data void, clinicians and policy makers still need to make the best decisions possible for current patients. As an alternative to making an educated guess about the benefit of adjuvant therapy, the study by Wang et al 21 attempts to offer quantitative, individualized survival predictions on the basis of the experience of more than 1,100 patients with resected gallbladder can- cer in the Surveillance, Epidemiology, and End Results–Medicare linked databases. Although one must acknowledge that models that are based on health claims data of the type found in the Surveil- lance, Epidemiology, and End Results–Medicare database may lack important clinical variables, and that models that are based on observational data may reflect selection biases, imperfect informa- tion is sometimes better than no information at all. In addition to addressing critiques of a previous model of adjuvant radiation for gallbladder cancer, the current chemoradiotherapy prediction model provides concrete adjuvant chemoradiotherapy survival benefit esti- mates on the basis of patient characteristics. 26-29 The Internet-based nomogram that is built on these results provides an interactive tool that may help patients, clinicians, and policy makers to make more informed, real-time decisions. 30 Although additional research would be needed to validate the predictions of the gallbladder cancer adjuvant chemoradiotherapy model described by Wang et al, 21 examples in the literature demon- strate the potential power of computer modeling, especially compre- hensive microsimulation models such as the Lung Cancer Policy Model (LCPM). 17 The LCPM was initiated a decade before the recent publication of the National Lung Screening Trial (NLST) results. Nonetheless, in contrast to two large previous clinical studies with widely divergent findings for computed tomography (CT) screening of individuals at high risk for lung cancer, the previously published LCPM results are remarkably consistent with the current NLST find- ings: a 6.7% reduction in all-cause mortality in the clinical trial of three annual CT screenings and a 4% reduction in all-cause mortality at 6 years in the LCPM analysis of five annual CT screenings. 17,20,31,33-37 This consistency in the magnitude of benefit is not a coincidence but rather is the result of a comprehensive microsimulation model of lung cancer development, progression, detection, treatment, and survival that accounts for competing mortality risks related to smoking and benign nodules and predicts the stage-shift effect of screening. The LCPM was extensively calibrated and validated with data from a vari- ety of sources. Simulating the NLST trial design and participants will provide an additional opportunity to validate the precision and accu- racy of model predictions. A model like the LCPM does not replace randomized controlled trials such as the NLST, but models can uniquely extend the time horizon and expand the population studied, JOURNAL OF CLINICAL ONCOLOGY E D I T O R I A L S © 2011 by American Society of Clinical Oncology 1 Journal of Clinical Oncology, Vol 29, 2011 http://jco.ascopubs.org/cgi/doi/10.1200/JCO.2011.37.8604 The latest version is at Published Ahead of Print on November 7, 2011 as 10.1200/JCO.2011.37.8604 Copyright 2011 by American Society of Clinical Oncology 2011 from 137.53.32.65 Information downloaded from jco.ascopubs.org and provided by at Oregon Health & Science University on November 7, Copyright © 2011 American Society of Clinical Oncology. All rights reserved.

JCO_Editorial_Nov2011

Embed Size (px)

DESCRIPTION

 

Citation preview

Page 1: JCO_Editorial_Nov2011

When a Decision Must Be Made: Role of ComputerModeling in Clinical Cancer ResearchRebecca A. Miksad, Beth Israel Deaconess Medical Center, Harvard Medical School, Boston, MA

See accompanying article doi: 10.1200/JCO.2010.33.8020

Every day, multidisciplinary oncology teams make dozens of treat-ment decisions that may have a tremendous impact on a patient’ssurvival and quality of life. Made with the best of intentions, thesedecisions are informed by basic science and clinical research findings,clinical experience, and health policy. All too often, results from thegold standard of clinical trial research, a randomized controlled trial(RCT), that fit the specific details of the patient’s situation are notavailable to guide these decisions.

Although this data gap occurs at times for all cancers, it is aconstant limitation for less common and biologically heterogeneousdiseases. For these cancers, such as those of the biliary tract, practicaltime and expense limitations restrict the number and combinations oftherapeutic strategies evaluated, the follow-up duration, and the pop-ulations studied in RCTs.1 And even in the most common cancers, thecostly failure of multiple trials that involve thousands of patients tomove cancer care forward has raised the need for alternate re-search paradigms.2-9

Enter computer modeling as a method to bridge currentknowledge gaps and to advance cancer clinical care and research.When performed correctly—rigorously developed, calibrated, andvalidated— computer modeling can maximize the informationthat is gained from current clinical, basic science, and epidemio-logic research efforts to facilitate informed clinical and healthpolicy decisions.10 This power stems from the ability of computermodels to produce novel comparative effectiveness findings, extendtrial results to longer time horizons, expand study findings to newpopulations, and refine expected outcomes. Last, but not least, com-puter models may also help differentiate between those scientific andclinical questions for which an RCT would be preferred but is not vitalfor decision making, and those questions for which the expense, time,and patient effort of an RCT is absolutely required to improve out-comes and to guide treatment and policy decisions.11-20

In the article that accompanies this editorial, Wang et al21 usedsurvival model techniques to predict the benefit of adjuvant chemo-therapy and chemoradiotherapy for patients with resected gallbladdercancer. Although the prognosis for these patients is usually grim andthe need for an effective treatment is great, there is a paucity of pub-lished information to guide adjuvant therapy choices.22-25 However,despite this data void, clinicians and policy makers still need to makethe best decisions possible for current patients.

As an alternative to making an educated guess about the benefitof adjuvant therapy, the study by Wang et al21 attempts to offer

quantitative, individualized survival predictions on the basis of theexperience of more than 1,100 patients with resected gallbladder can-cer in the Surveillance, Epidemiology, and End Results–Medicarelinked databases. Although one must acknowledge that models thatare based on health claims data of the type found in the Surveil-lance, Epidemiology, and End Results–Medicare database may lackimportant clinical variables, and that models that are based onobservational data may reflect selection biases, imperfect informa-tion is sometimes better than no information at all. In addition toaddressing critiques of a previous model of adjuvant radiation forgallbladder cancer, the current chemoradiotherapy prediction modelprovides concrete adjuvant chemoradiotherapy survival benefit esti-mates on the basis of patient characteristics.26-29 The Internet-basednomogram that is built on these results provides an interactive toolthat may help patients, clinicians, and policy makers to make moreinformed, real-time decisions.30

Although additional research would be needed to validate thepredictions of the gallbladder cancer adjuvant chemoradiotherapymodel described by Wang et al,21 examples in the literature demon-strate the potential power of computer modeling, especially compre-hensive microsimulation models such as the Lung Cancer PolicyModel (LCPM).17 The LCPM was initiated a decade before the recentpublication of the National Lung Screening Trial (NLST) results.Nonetheless, in contrast to two large previous clinical studies withwidely divergent findings for computed tomography (CT) screeningof individuals at high risk for lung cancer, the previously publishedLCPM results are remarkably consistent with the current NLST find-ings: a 6.7% reduction in all-cause mortality in the clinical trial of threeannual CT screenings and a 4% reduction in all-cause mortality at 6years in the LCPM analysis of five annual CT screenings.17,20,31,33-37

This consistency in the magnitude of benefit is not a coincidence butrather is the result of a comprehensive microsimulation model of lungcancer development, progression, detection, treatment, and survivalthat accounts for competing mortality risks related to smoking andbenign nodules and predicts the stage-shift effect of screening. TheLCPM was extensively calibrated and validated with data from a vari-ety of sources. Simulating the NLST trial design and participants willprovide an additional opportunity to validate the precision and accu-racy of model predictions. A model like the LCPM does not replacerandomized controlled trials such as the NLST, but models canuniquely extend the time horizon and expand the population studied,

JOURNAL OF CLINICAL ONCOLOGY E D I T O R I A L S

© 2011 by American Society of Clinical Oncology 1Journal of Clinical Oncology, Vol 29, 2011

http://jco.ascopubs.org/cgi/doi/10.1200/JCO.2011.37.8604The latest version is at Published Ahead of Print on November 7, 2011 as 10.1200/JCO.2011.37.8604

Copyright 2011 by American Society of Clinical Oncology2011 from 137.53.32.65

Information downloaded from jco.ascopubs.org and provided by at Oregon Health & Science University on November 7,Copyright © 2011 American Society of Clinical Oncology. All rights reserved.

Page 2: JCO_Editorial_Nov2011

evaluate alternative screening strategies, and assess the relative value ofpotential policy interventions.

These unique abilities of computer models are particularly valu-able when policy decisions must be made on the basis of available data,including observational or single-arm studies, until more definitiveRCT results are available and in situations in which funding, time, andpatient populations limit answerable research questions.32 For exam-ple, a simulated control group for the single-arm Mayo lung cancerscreening study with the LCPM allowed exploration of the contradic-tory results of two large clinical studies.33-37 In addition, the LCPMwas able to assess 15-year survival estimates for three different screen-ing strategies—a significantly longer time-line and a more complexanalysis than is typically possible in an RCT.20

Simulationmodelingalongsideclinicalresearchmayalsostrengthentrial findings by allowing the exploration of areas of potential bias:concerns about the NSLT false-positive rate and potential for overdi-agnosis that were suggested by the 16-year results of the Mayo studycan be explicitly evaluated in a model.32 In addition, the flexibility ofthe simulation models also allows evaluation of questions raised by theNSLT study that are important to clinicians and policy makers but forwhich repeated, long-term clinical trials are not possible: the benefitof screening in populations with lower or variable adherence, the effectof extending annual screening beyond 3 years, as well as the impact ofscreening on light smokers and genomic subgroups. On the societallevel, an additional value of modeling is to generate novel hypothesesand to identify research questions that merit clinical trial resources.

Although it is not a comprehensive microsimulation model likethe LCPM, the model described by Wang et al21 moves gallbladdercancer clinical care forward by offering evidence in favor of adjuvantchemoradiotherapy for some patients and by providing guidance forgallbladder cancer research. For example, future studies can take ad-vantage of model efficacy estimates to help guide clinical trial de-sign, to identify subgroups (adjuvant chemoradiotherapy benefitmay be small for patients with node-negative disease), to helprefine target populations, and to highlight areas for additionalresearch (the interaction between extended lymphadenectomy andadjuvant chemotherapy). Building from the results by Wang et al,a microsimulation model calibrated to and validated with externaldata sets may increase the robustness and precision of adjuvant che-moradiotherapy model predictions.

The statistical aspects of the survival analysis model by Wang etal,21 the majority of which was originally reported in a technical jour-nal, also merit discussion.38 Although the Cox proportional hazardratio model is commonly used in medicine, other survival analysismethods, such as the accelerated failure time log normal model usedby Wang et al, may more appropriately reflect the biology of somecancer scenarios.39 For example, accelerated failure time models allowthe intervention effect to change over time, as is seen when the effec-tiveness of chemotherapy decreases over time because of resistance.Similar to findings in other cancers,40,41 Wang et al demonstrate theneed for careful consideration of the best analytic method by docu-menting performance variations for five gallbladder survival modelanalysis approaches. The oncology community should expect rigor-ous consideration of the appropriate survival analysis methods in allclinical trial and modeling research. As the application of cancer mod-els expands, researchers, clinicians, and policy makers who under-stand both the clinical research and computer modeling worlds areneeded to translate between model results, clinical trial design, clinical

practice, and health policy to ensure that the best decisions are madefor patients.

AUTHOR’S DISCLOSURES OF POTENTIAL CONFLICTS OF INTERESTThe author(s) indicated no potential conflicts of interest.

REFERENCES1. Knudsen AB, McMahon PM, Gazelle GS: Use of modeling to evaluate the

cost-effectiveness of cancer screening programs. J Clin Oncol 25:203-208, 20072. Weiser MR: Rectal cancer trials: No movement. J Clin Oncol 29:2746-

2748, 20113. Miksad RA: Pathologic complete response and toxicity results from the

STAR-01 Trial evaluating the addition of oxaliplatin to neodjuvant chemoradiationfor locally advanced rectal cancer. J Clin Oncol 29, 2011. http://jco.ascopubs.org/content/29/20/2773/suppl/DC2

4. LoRusso PM, Schnipper LE, Stewart DJ, et al: Translating clinical trials intomeaningful outcomes. Clin Cancer Res 16:5951-5955, 2010

5. LoRusso PM, Anderson AB, Boerner SA, et al: Making the investigationaloncology pipeline more efficient and effective: Are we headed in the rightdirection? Clin Cancer Res 16:5956-5962, 2010

6. Schnipper LE, Meropol NJ, Brock DW: Value and cancer care: Toward anequitable future. Clin Cancer Res 16:6004-6008, 2010

7. Miksad RA, Schnipper L, Goldstein M: Does a statistically significantsurvival benefit of erlotinib plus gemcitabine for advanced pancreatic cancertranslate into clinical significance and value? J Clin Oncol 25:4506-4507, 2007;author reply 4508

8. Butler D: Translational research: Crossing the valley of death. Nature453:840-842, 2008

9. Booth CM: Evaluating patient-centered outcomes in the randomizedcontrolled trial and beyond: Informing the future with lessons from the past. ClinCancer Res 16:5963-5971, 2010

10. Rutter CM, Knudsen AB, Pandharipande PV: Computer disease simulationmodels: Integrating evidence for health policy. Acad Radiol 18:1077-1086, 2010

11. Zauber AG, Lansdorp-Vogelaar I, Knudsen AB, et al: Evaluating TestStrategies for Colorectal Cancer Screening: Age to Begin, Age to Stop, andTiming of Screening Intervals—A Decision Analysis of Colorectal Cancer Screen-ing for the U.S. Preventive Surveillance Modeling Network (CISNET). Rockville,MD, Agency for Healthcare Research and Quality, 2009

12. Knudsen AB, Lansdorp-Vogelaar I, Rutter CM, et al: Cost-effectiveness ofcomputed tomographic colonography screening for colorectal cancer in themedicare population. J Natl Cancer Inst 102:1238-1252, 2010

13. Pandharipande PV, Choy G, del Carmen MG, et al: MRI and PET/CT fortriaging stage IB clinically operable cervical cancer to appropriate therapy:Decision analysis to assess patient outcomes. AJR Am J Roentgenol 192:802-814, 2009

14. Pandharipande PV, Gervais DA, Hartman RI, et al: Renal mass biopsy toguide treatment decisions for small incidental renal tumors: A cost-effectivenessanalysis. Radiology 256:836-846, 2010

15. Ladapo JA, Jaffer FA, Hoffmann U, et al: Clinical outcomes and cost-effectiveness of coronary computed tomography angiography in the evaluation ofpatients with chest pain. J Am Coll Cardiol 54:2409-2422, 2009

16. Pandharipande PV, Harisinghani MG, Ozanne EM, et al: Staging MRlymphangiography of the axilla for early breast cancer: Cost-effectiveness anal-ysis. AJR Am J Roentgenol 191:1308-1319, 2008

17. McMahon PM, Kong CY, Johnson BE, et al: Estimating long-term effec-tiveness of lung cancer screening in the Mayo CT screening study. Radiology248:278-287, 2008

18. Gazelle GS, Hunink MG, Kuntz KM, et al: Cost-effectiveness of hepaticmetastasectomy in patients with metastatic colorectal carcinoma: A state-transition Monte Carlo decision analysis. Ann Surg 237:544-555, 2003

19. Huang ES, Gazelle GS, Hur C: Consensus guidelines in the management ofbranch duct intraductal papillary mucinous neoplasm: A cost-effectiveness anal-ysis. Dig Dis Sci 55:852-860, 2010

20. McMahon PM, Kong CY, Weinstein MC, et al: Adopting helical CTscreening for lung cancer: Potential health consequences during a 15-year period.Cancer 113:3440-3449, 2008

21. Wang SJ, Lemieux A, Kalpathy-Cramer J, et al: Nomogram for predictingthe benefit of adjuvant chemoradiotherapy for resected gallbladder cancer. J ClinOncol doi: 10.1200/JCO.2010.33.8020

22. Macdonald OK, Crane CH: Palliative and postoperative radiotherapy inbiliary tract cancer. Surg Oncol Clin N Am 11:941-954, 2002

23. Southwest Oncology Group: S0809–Phase II: A Phase II Trial of AdjuvantCapecitabine/Gemcitabine Chemotherapy Followed by Concurrent Capecitabine

Rebecca A. Miksad

2 © 2011 by American Society of Clinical Oncology JOURNAL OF CLINICAL ONCOLOGY

2011 from 137.53.32.65Information downloaded from jco.ascopubs.org and provided by at Oregon Health & Science University on November 7,

Copyright © 2011 American Society of Clinical Oncology. All rights reserved.

Page 3: JCO_Editorial_Nov2011

and Radiotherapy in Extrahepatic Cholangiocarcinoma [protocol abstract]. http://swog.org/Visitors/ViewProtocolDetails.asp?ProtocolNumber�S0809

24. Jarnagin WR, Ruo L, Little SA, et al: Patterns of initial disease recurrenceafter resection of gallbladder carcinoma and hilar cholangiocarcinoma: Implica-tions for adjuvant therapeutic strategies. Cancer 98:1689-1700, 2003

25. National Comprehensive Cancer Network: NCCN Guidelines: HepatobiliaryCancers, Version 2.2011. http://www.nccn.org/professionals/physician_gls/f_guidelines.asp

26. Yu JB, Zelterman D, Decker RH, et al: Impact of immediate postoperativedeath on the estimation of a survival benefit from postoperative radiation therapyfor cancer of the gallbladder. J Clin Oncol 26:4523, 2008; author reply 4524-4526

27. Cleary SP, Tan JC, Law CH, et al: Treatment considerations for gallbladdercancer should include extent of surgery. J Clin Oncol 26:4521-4522, 2008; authorreply 4524-4526

28. Arroyo GF, Lemoine G: Prediction model for adjuvant radiation therapy forgallbladder cancer: Not ready to be used. J Clin Oncol 26:4522-4523, 2008;author reply 4524-4526

29. Wang SJ, Fuller CD, Kim JS, et al: Prediction model for estimating thesurvival benefit of adjuvant radiotherapy for gallbladder cancer. J Clin Oncol26:2112-2117, 2008

30. Knight Cancer Institute of Oregon Health and Science University: GallbladderCancer Adjuvant Therapy, 2011. http://skynet.ohsu.edu/nomograms/gallbladder/

31. National Lunch Screening Trial Research Team, Aberle DR, Adams AM, etal: Reduced lung-cancer mortality with low-dose computed tomographic screen-ing. N Engl J Med 365:395-409, 2010

32. Sox HC: Better evidence about screening for lung cancer. N Engl J MedAug 365:455-457, 2011

33. Henschke CI, Naidich DP, Yankelevitz DF, et al: Early lung cancer actionproject: Initial findings on repeat screenings. Cancer 92:153-159, 2001

34. Swensen SJ, Jett JR, Hartman TE, et al: CT screening for lung cancer:Five-year prospective experience. Radiology 235:259-265, 2005

35. Swensen SJ, Jett JR, Sloan JA, et al: Screening for lung cancer withlow-dose spiral computed tomography. Am J Respir Crit Care Med 165:508-513,2002

36. International Early Lung Cancer Action Program Investigators, HenschkeCI, Yankelevitz DF, et al: Survival of patients with stage I lung cancer detected onCT screening. N Engl J Med 355:1763-1771, 2006

37. Bach PB, Jett JR, Pastorino U, et al: Computed tomography screening andlung cancer outcomes. JAMA 297:953-961, 2007

38. Wang SJ, Kalpathy-Cramer J, Kim JS, et al: Parametric survival models forpredicting the benefit of adjuvant chemoradiotherapy in gallbladder cancer. AMIAAnnu Symp Proc 2010:847-851, 2010

39. Ahmed FE, Vos PW, Holbert D: Modeling survival in colon cancer: Amethodological review. Mol Cancer 6:15, 2007

40. Anderson JR, Cain KC, Gelber RD, et al: Analysis and interpretation of thecomparison of survival by treatment outcome variables in cancer clinical trials.Cancer Treat Rep 69:1139-1146, 1985

41. Smith LK, Lambert PC, Botha JL, et al: Providing more up-to-date esti-mates of patient survival: A comparison of standard survival analysis with periodanalysis using life-table methods and proportional hazards models. J Clin Epide-miol 57:14-20, 2004

DOI: 10.1200/JCO.2011.37.8604; published online ahead of print atwww.jco.org on November 7, 2011

■ ■ ■

Acknowledgment

R.A.M. is supported by the National Cancer Institute Grant No. 1 K23 CA139005-01A1.

Editorials

www.jco.org © 2011 by American Society of Clinical Oncology 3

2011 from 137.53.32.65Information downloaded from jco.ascopubs.org and provided by at Oregon Health & Science University on November 7,

Copyright © 2011 American Society of Clinical Oncology. All rights reserved.