Multiple Sclerosis and Related Disorders (2013) 2, 377–384
Available online at www.sciencedirect.com
journal homepage: www.elsevier.com/locate/msard
An agent-based simulation model for informed shared decision making in multiple sclerosis Mário Veloson,1 Hospital Egas Moniz (Centro Hospitalar de Lisboa Ocidental, EPE), Rua da Junqueira, 126, 1349-019 Lisboa, Portugal Received 11 February 2013; received in revised form 25 March 2013; accepted 7 April 2013
KEYWORDS
Abstract
Multiple sclerosis; Shared decision making; Simulation model; Prognosis; Decision aids; Agent-based system
Shared decision making (SDM) is concerned with patient involvement into medical decisions and chronic conditions such as Multiple sclerosis (MS), with only partially effective treatments leading to potential severe side effects, conflicting evidence, and uncertain evidence on outcomes, constitute a typical condition for SDM. As treatment options increase and patients participate more intensively in decisions, the need for evidence-based information (EBI) becomes clear. Natural history (NH) studies of MS represent the basic sources for required EBI and are especially useful to contribute to the practical exercise of prognosis formulation and to enable the evaluation of effectiveness in the context of treatment. Several of these identify early clinical factors predictive of the course of MS but there is no consensus method for determining the long term progression of disability and evolution of individual patients on the basis of observations on the early stages of the disease, which constitutes a major challenge for the practicing neurologist. Aiming at delivering more reliable prognosis estimation, this study combines the distribution of patients reaching specific levels of disability within defined time periods as determined in NH studies, with disability curves and severity scores as a function of time, in terms of percentiles and deciles respectively, derived from longitudinal data analysis studies. A computer agent-based simulation model was implemented as a comprehensive and easy to utilize tool able to predict and monitor progression of disability in MS patients, and to support the neurologist discussing prognosis scenarios with the individual patient for effective SDM. & 2013 Elsevier B.V. All rights reserved.
1.
n
Tel.: +351 964020950 (Mob); fax: +351 217993913. E-mail addresses: [email protected], [email protected]. 1 Present address: ARN—Anestesia, Reanimação e Neurologia, Lda, Campo Grande 14-6A, 1700-092 Lisboa, Portugal. 2211-0348/$ - see front matter & 2013 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.msard.2013.04.001
Introduction
The goal of both diagnosis and therapeutic decisions is to improve the prognosis for the patient, since prognosis refers to all medical outcomes that may occur during the patient's disease process. This also implies that patient management should be driven by two major prognosis related topics: the natural history (NH), which is the prognosis of the disease without medical interventions, and the prognosis changes as
378 resulting from medical interventions. Additionally, prognosis is a major concern to the patient who wants to be informed about his prospects and prognosis related information is a mandatory requirement for an informed and active participation of the patient on self-clinical decision-making. A major challenge in multiple sclerosis (MS) for the practicing neurologist is to make a prediction of the long term evolution of individual patients on the basis of observations on the early stages of the disease. An immediate effect of such difficulty is to inhibit the communication to the patient of a realistic estimation of his/her evolution, particularly in a long term basis. Prognosis' estimates or prediction can be made in several ways. As opposed to an informal way (e.g. intuitively or using expert opinions), modern patient management requires that appropriateness of medical interventions is supported by scientific evidence, integrating clinical expertise, patient values, and the best research evidence into the decision making process for patient care, the basis for evidencebased medicine (Sackett et al., 1996). Including patient values and preferences in the clinical decision-making is an ethical issue (Hughes and Larson, 1991) and can contribute to improve patient care (O′Connor et al., 2007; Sepucha et al., 2004). Shared decision-making (SDM) is a major component of such patient-centered care, and has been defined as a process that “allows both physicians and patients to honor the values and preferences of the patient, while also permitting the physician to provide medical expertise to promote the patient′s health” (O′Connor et al., 2007). Hence, SDM takes informed consent a step further in the process of communication between a patient and a physician. It is much more than obtaining the patient′s authorization or agreement to undergo a specific medical intervention; SDM means to involve the patient actively on the medical decisions and such patient empowerment also means giving the patient more responsibility. Chronic conditions such as MS, with only partially effective treatments with potential severe side effects, conflicting evidence, and uncertain evidence on outcomes, where the benefit–harm ratio is short and/or doubtful, or when available options have different benefit–harm profiles that patients value differently, constitute a typical condition for this SDM and enforces its need (Heesen et al., 2011; Wennberg et al., 2002). As treatment options increase and patients participate more intensively in decisions, the need for evidence-based information becomes clear. There is an evidence that patients may, in exchange for therapeutic benefits, be willing to accept greater levels of risk than are actually posed by some therapies (Calfee, 2006; Johnson et al., 2007a, 2007b). Drug therapies for MS offer a range of potential benefits, but they may also involve lifethreatening risks, including liver failure, leukemia, and progressive multifocal leukoencephalopathy (Brassat et al., 2002; Francis et al., 2003; Yousry et al., 2006). Yet patients may be misinformed particularly by means of internet information, misinterpret the results of scientific research (Jadad et al., 2000; Kaplan and Brennan, 2001), have unrealistic expectations of treatment benefits and harms, and clinicians may be poor judges of a patient′s values (O′Connor et al., 2007). There also is evidence that patient decision aids are better than usual care in improving patients' knowledge and
M. Veloso expectations about interventions, as well as improving agreement between values and choice (Elwyn et al., 2006). A diversity of methods has emerged for decision support from different scientific fields such as Statistics, Decision Analysis and Artificial Intelligence. Cox models, recursive partitioning analysis, Weibull models, decision trees, Markov models, Partially Observable Markov Decision Process, Bayesian networks and influence diagrams. No matter the approach, the first step is to build up a prognostic model able to predict the probability of some outcome as optimally as possible. Several studies have been performed to identify early clinical factors predictive of the MS course (Levic et al., 1999; Runmarker and Andersen, 1993; Weinshenker et al., 1989, 1991; Kremenchutzky et al., 2006; Confavreux and Vukusic, 2006; Ebers, 2005; Vukusic and Confavreux, 2007; Scalfari et al., 2012), such as: gender, disease course, age at onset of disease, initial symptoms, number of functional systems involved, first interval attack, attack frequency, and incomplete remission after the first episode. However, the majority of work in this area is not focused on the individual prognosis to the patient; neither does address treatment effects on the NH of the disease. Daumer et al. (2007), describe an online analytical processing tool that matches the characteristics of a given patient with the most similar patients of the Sylvia Lawry Centre for Multiple Sclerosis Research database. An “individual risk profile” in terms of the disease course of all similar patients in the database is displayed, hence enabling to project a hypothesized outcome for that patient (Daumer et al., 2007). The main limitations of this tool are related to the characteristics of the patients included in the database. The clinical data are derived only from the placebo groups of randomized clinical trials, and the respective observation period is limited to a maximum of three years (Daumer et al., 2007). In the study of Wolfson and Confavreux (1985, 1987), a Markov model is proposed to represent the disease course by means of transitions between disease states, as to evaluate the effect of prognostic factors on those transitions. Because Markov processes are memory less, once a state is known, the future evolution of the disease is independent of the past evolution. This limitation is handled in the work of Bergamaschi et al. (2007, 2001), by proposing a Bayesian model specifying the full joint probability distribution for a set of random variables that characterize the entire course of the disease. The risk of reaching secondary progression was significantly related to specific clinical factors presented during the first year of the disease, all of them associated with a specific statistical weight, the Bayesian local relative risk, used to calculate the Bayesian Risk Estimate for MS (BREMS) score for any given patient (Bergamaschi et al., 2007, 2001). However, no other prognostic outcomes are provided. Achiron et al. (2003) use NH information from a large database to generate longitudinal disability curves for prediction of disease progression based on the mean Expanded Disability Status Scale (EDSS) scores from the first year of disease onset represented as a major percentile group. These curves represent cohort percentiles and enable to foresee the relative risk of disease progression, as well as to identify deviations in the curves (Achiron
An agent-based simulation model for informed shared decision making in multiple sclerosis et al., 2003). Similarly, the global Multiple Sclerosis Severity Score (MSSS) (Roxburgh et al., 2005) has been proposed as a population disability assessment tool enabling comparisons of relative disease severity at all EDSS levels for a given disease duration. A more recent study (Gray et al., 2008) confirmed the validity of the MSSS as a 5-year severity rank predictor in individual patients. In spite of the increasing number of prognostic models developed in Medicine, the fact is that such prognostic models are seldom used in the clinical practice (Wyatt and Altman, 1995; Dong et al., 2012). A major challenge, often underestimated, consists of the ability to provide the relevant information for decision-making in a comprehensive
Fig. 1
379
and easy to process format both for patients and physicians. Different patients have different information needs, and different ways of decision-making (Mazur and Hickam, 1991). Moreover, many people, including physicians, have difficulty in process critically quantitative information (Woloshin et al., 2001). Considering that most information relevant to decision-making is inherently probabilistic, new ways for presenting such information are required (Edwards et al., 2002; Heesen et al., 2011). Building and using models is part of everybody′s life, being simplification and abstraction of the real system the key in modeling. Agent-based modeling (ABM) is a type of modeling in which the focus is on representing agents (such
Simulation model user interface.
380
M. Veloso
as people) and their interactions (Miller and Page, 2007), enabling to effectively capture a very rich set of complex behaviors and interactions, hence highly suited to modeling complex phenomena. ABM capabilities explain its extensive use in a diversity of fields, including decision-support and Medicine (Cook et al., 2011; Small, 2007). The main goal of the present work is to provide the clinician with an easy to use simulation tool concerning individual long term (30 years) disability prediction and treatment effect on the estimated individual prognosis, in order to enable a comprehensive interaction with the patient with MS for effective SDM.
2. 2.1.
Methods The simulation model
The simulation model was implemented using the Netlogo 5.0.2 programmable modeling environment (Wilensky and NetLogo, 1999). Netlogo is a programming language with predominantly agent-oriented attributes, with unique capabilities that make it extremely powerful for producing and visualizing simulations of multi-agents systems. Taking into account that available disease-modifying agents are only indicated for the relapsing-remitting form of MS, the proposed simulation model only addresses this form of disease. For simulation purposes the model uses virtual population of 100 agents (simulating persons with relapsing-remitting MS) and all calculations are reflected in this population by means of different sorts of graphics (Fig. 1). Basically, disease progression is derived from the number of years from disease onset to EDSS scores 4, 6 and 7, according to published data of the Lyon database (Confavreux and Compston, 2006; Confavreux et al., 2003, 2000; Confavreux and Vukusic, 2006). The 95% confidence interval of the time to assignment of the different EDSS scores was used as reference for building the state transition rules that guarantee a proper probability distribution of the 100 agents (virtual MS patients) along the time. Transitions intervals are anticipated or delayed in accordance to the individual patient characteristics as entered by the user. These characteristics are the clinical variables used to build up the BREMS score (Bergamaschi et al., 2007, 2001), corresponding to known prognostic factors: age at onset of disease, gender, sphincter onset, pure motor onset, motorsensory onset, sequel after onset, number of involved functional systems at onset, number of sphincter plus motor relapses, EDSS≥4. Dependent on selected patient characteristics, the system will display a 30 years prediction of the probabilities to be at EDSS scores o4, 4, 6 and 7 in two graphics of a group of 100 patients with similar clinical characteristics. One graphic, will display 100 “persons” with different color intensity (pink for females, blue for males) representing the different EDSS possibilities. The second graphic, will present the yearly distribution of the different EDSS states along the 30 years projection. The chance of being in a progressive phase of the disease at 10 years after onset, and the value of patient quality adjusted life years (QALYs) after 30 years will be also displayed numerically in two distinct
boxes. Patient QALYs are calculated as a function of the number of years at the different EDSS states estimated for the individual patient. By providing the system with the EDSS in subsequent years, the prognosis estimation in terms of disability progression will be refined. The patient EDSS score will be graphically situated in relation with the 10 years disability progression curves representing the respective annually percentile figures (Achiron et al., 2003). On the other hand, the yearly introduced EDSSs will enable to calculate the MSSS for those years, hence to make a 30 years projection corresponding to the maximum EDSS score expected to the derived decile (Roxburgh et al., 2005). Besides the predictive facilities, these graphics will enable the user to monitor disability progression along the years, hence to early recognize any up-deviation which certainly correspond to a clinical deterioration. The simulation model enables three ways for treatment assessment. Firstly, pressing the “assess” button after selecting a treatment from the list, a new 30 years prediction is made by the system reflecting the treatment effect in terms of “Absolute Risk Reduction” on two variables in accordance to published data in the literature: relapse free and progression free at 2 years treatment (Filippini et al., 2003; Freedman et al., 2008; Jacobs et al., 1996; Johnson et al., 1995; Polman et al., 2006). Treatment effect is calculated through the impact of those two factors upon the time to reach the different EDSS states, and this will be reflected in the estimation of the EDSS distribution along the 30 years that will be displayed, together a prediction of the QALYs gain at 30 years, Table 1 Validation group as compared to overall patient characteristics.
Gender Male—N (%) Female—N (%)
RRMSa (N =127)
(N =50)
PPMSmb (N =9)
31 (24.4%) 96 (75.6%)
13 (26%) 37 (74%)
5 (55.6%) 4 (44.4%)
Age at disease onset Mean (SDc) 28.0 (7 17.4) Median 27.8
28.42 (7 8.39) 25.65
Functional system affected—N (%) Motor 33 (26.0%) 12 (24.0%) Sensory 37 (29.1%) 18 (36.0%) Coordenation 7 (5.5%) 1 (2.0%) Visual 34 (26.8%) 14 (28.0%) Brainstem 11 (8.7%) 5 (10.0%) Undefined 5 (3.9%) 0 Interval 1st–2nd relapse 4.8 (711.1) 4.66 Mean (SDc) (76.03) Median 2.0 2.59 a
RRMS: relapsing-remitting multiple sclerosis. PPMS: primary progressive multiple sclerosis. c SD: standard deviation. b
An agent-based simulation model for informed shared decision making in multiple sclerosis estimated as described before. Secondly, treatment effect can be monitored through the early recognition of any deviation in the disability curves and MSSS progression graphics. Finally, by introducing yearly data in addition to the EDDS (number of mild and severe relapses, and number of new Magnetic Ressonance Imaging lesions), a textual recommendation is displayed in accordance to published consensus adopted criteria (Jeffery et al., 2002).
2.2.
Model validation
Data for the validation of the simulation model was retrospectively obtained from the hospital medical records of a group of 50 patients selected from the total of the 173 patients observed by the author between January and July of 2012 in the MS outpatient clinic of Hospital Egas Moniz. The following criteria were used for selecting the validation group: Patients with relapsing-remitting multiple sclerosis (RRMS) with at least 10 years of disease evolution and sufficient data in their medical records, namely the number of relapses at years 1, 2, 5 and 10, and the EDSS scores at years 2, 5 and 10. The study has been approved by the Hospital Ethics Committee and performed in accordance with the ethical standards laid down in the 1964 Declaration of Helsinki and its later amendments. The validation procedure consisted of assessing the concordance degree between the simulation model predictions and the real value presented by the patients, for 10 and 20 years estimations, using patient EDSS at years 2 and 5 of disease evolution.
3.
Results
Thirty seven patients among the 173 patients observed were discharged because of lack of data in their clinical record together with patient incapacity in describing the facts related to their initial clinical history. Data from the validation group of 50 patients is presented in Table 1, as compared to the data of the overall 136 patients. All the data and validation procedures described next focus on this subgroup of 50 patients. The mean age at onset was 28.4 years (7 8.4, median 25,6), mean disease evolution 17.1 (77.8), and the mean EDSS at the last observation was 3 (7 2.2), median 2. The number of relapses and EDSS scores were recorded (Table 2); the mean time to reach EDSS 4 was 12.4 (76.9) years and EDSS 6 was 20.9 (78.8), median 11 and 18 respectively.
Table 2
b
Table 3 presents the correlation between the number of relapses and EDSS at years 5 and 10, while Table 4 presents the correlation between EDSS scores at different years. As to the model 10 years prediction, a strong correlation was found both with the disability curves (Pearson coefficient= 0.75; po0.0001) and the MSSS projection (Pearson coefficient = 0.67; po0.0001). If one considers the minimum–maximum interval of MSSS estimation, in 76% of cases the system made an adequate prediction. As to the 20 years prediction of the MSSS, similar correlation degree (Pearson coefficient = 0.67; p= 0.0016) and probability of min–max correct estimation (76%—13 of 17 patients) were found. Because lack of data it was not possible to evaluate the model behavior concerning the patient characteristics at disease onset, neither treatment effect.
4.
Discussion
NH studies of MS are especially useful to contribute to the practical exercise of prognosis formulation and to enable the evaluation of effectiveness in the context of treatment. Analysis of the MS' NH using different statistical approaches, proposed similar conclusions concerning the prognosis influence of some basic predictors. However, because of high variation amongst patients' disease course only rough predictions for the individual patient are possible. Table 3 Correlation between number of relapses and EDSS.
Number of relapses at year 2 Number of relapses at year 5 a
EDSSa at year 5
EDSSa at year 10
0.50 (p =0.0001) 0.60 (po0.0001)
0.26 (p =0.031) 0.40 (p =0.002)
EDSS: Expanded Disability Status Scale.
Table 4 Correlation between EDSSa at early stages with EDSSa at 10 years.
EDSSa at year 2 EDSSa at year 5 a
EDSSa at year 5
EDSSa at year 10
0.75 (po0.0001) –
0.60 (po0.0001) 0.77 (po0.0001)
EDSS: Expanded Disability Status Scale.
Recorded relapses and EDSSa scores.
Mean number of relapses (SDb) EDSS mean score (SDb) a
381
EDSS: Expanded Disability Status Scale. SD: standard deviation.
Year 1
Year 2
Year 5
Year 10
0.47 (70,7) –
0.80 (71.1) 0.76 (71.0)
1.60 (71.4) 1.27 (71.0)
2,76 (72.0) 2,02 (72.0)
382 The present study combined the results of longitudinal studies (Achiron et al., 2003; Bergamaschi et al., 2007, 2001; Roxburgh et al., 2005) with previous NH studies and disease prognostic factors, in order to provide clinicians with an easy to use tool concerning individual long term disability prediction and treatment effect on the estimated individual prognosis, enabling a comprehensive interaction with the patient with MS for effective SDM. For such purpose, a computer model able to simulate individual patient prognosis was implemented. Based on patient characteristics at disease onset, the model provides individual 30 years disability prediction derived from the time to reach EDSS scores 4, 6 and 7. It also enables to asses treatment effect on the estimated individual prognosis, as well as to monitor deviations in the disability curves (percentiles) and in the MSSS (deciles), as patient data in subsequent years is provided. Modeling and simulation have been used in a variety of scientific domains (Law and Kelton, 1999), and significant improvement in decision-making, efficiency and quality reported (Kuljis et al., 2007). The main advantage of modeling is to facilitate to understand the behavior of a real system and then to test it through a variety of simulations of different scenarios. The simulation model now presented is drawn upon available scientific evidence relevant for the defined purpose. However, in general, available data are scarce and/or not adequate for the modeling requirements limiting model validation. Ultimately, the main limitations of the current model are the limitations of the data that the model proposes to represent and process. In order to ensure consistency, a single data source was used but a metaanalysis of several sources is also a valid option. Promising results were obtained through performed simulations with data from real patients when complementing model estimations based on patient characteristics at disease onset with initial defined percentiles and deciles. Using the initial EDSS it is already possible to predict progression along percentiles and deciles regardless of relapses, since relapses seem not significantly influence long term disability (Confavreux et al., 2000), as also documented in the present study. Initial estimations can be later refined by specifying EDSS scores in subsequent years. A major added value of the current model consists of its monitoring capabilities particularly by enabling the clinician to be aware early of any change to expected course of disease. Any up-deviation from assigned percentile or decile means unexpected disease deterioration, hence the need for intervention (e.g. treatment switch or escalation). This capability is further refined by means of the criteria for the identification of sub-optimal treatment response. The results obtained through the assessment of the different disease-modifying agents effect in terms of QALYs gain after 30 years deserve a comment. Whether the reduced gain obtained is a simulation problem or it represents actually a low impact of treatment into long term accumulated quality of life, remains a question to be solved. Nevertheless, lack of a major impact of disease-modifying agents upon disease severity as measured by the MSSS has been previously reported (Pachner and Steiner, 2009). Previewed future model developments are twofold. On one hand, improvement of the data related to NH of the
M. Veloso disease supporting the model is an ongoing process. In particular, partnership with other research and/or MS centers is being envisaged. The overall idea is to validate the model with different data sets and to evaluate model simulations as to the impact of patient characteristics (e.g. prognostic factors) at disease onset upon the time to reach the different EDSS states. Similarly, treatment effect on state transition intervals also needs deeper testing. On the other hand, model integration in a patient record system seems to be a natural extension to current achievements. NetLogo, besides enabling to delivering models as java applets, hence easily incorporated into web pages, also provides a programming interface to other programming languages.
5.
Conclusions
To be useful for clinicians, a MS prognosis model should provide a reliable prediction at the individual level and a way to monitor disability progression in MS patients. By combining two different approaches of modeling disability progression, the proposed model of prognosis prediction seems to be at least as good as the ones that can be extracted from available scientific evidence. Therefore, one can conclude that the current model is a valuable tool and, because of its conceptual simplicity and easy utilization, this type of simulation model can be used in everyday clinical practice both to monitor progression of disability, and to support the neurologist discussing prognosis scenarios with the individual patient for effective SDM.
Conflict of interest The author has no conflicts of interest or disclosure to report. This work has received no financial support.
References Achiron A, Barak Y, Rotstein Z. Longitudinal disability curves for predicting the course of relapsing-remitting multiple sclerosis. Multiple Sclerosis 2003;9:486–91. Bergamaschi R, Quaglini S, Trojano M, Amato MP, Tavazzi E, Paolicelli D, et al. Early prediction of the long term evolution of multiple sclerosis: the Bayesian Risk Estimate for Multiple Sclerosis (BREMS) score. Journal of Neurology, Neurosurgery, and Psychiatry 2007;78:757–9. Bergamaschi R, Berzuini C, Romani A, Cosi V. Predicting secondary progression in relapsing-remitting multiple sclerosis: a Bayesian analysis. Journal of the Neurological Sciences 2001;189:13–21. Brassat D, Recher C, Waubant E, Le Page E, Rigal-Huguet F, Laurent G, et al. Therapy-related acute myeloblastic leukemia after mitoxantrone treatment in a patient with MS. Neurology 2002;59:954–5. Calfee, J.E. A representative survey of MS patients on attitudes toward the benefits and risks of drug therapy. AEI-Brookings Joint Center for Regulatory Studies; 2006. Confavreux C, Compston A. The natural history of multiple sclerosis. In: Compston A, editor. McAlpine's multiple sclerosis. 4th ed. London: Churchill Livingstone Elsevier; 2006. p. 183–272. Confavreux C, Vukusic S, Adeleine P. Early clinical predictors and progression of irreversible disability in multiple sclerosis: an amnesic process. Brain 2003;126:770–82.
An agent-based simulation model for informed shared decision making in multiple sclerosis Confavreux C, Vukusic S, Moreau T, Adeleine P. Relapses and progression of disability in multiple sclerosis. New England Journal of Medicine 2000;343:1430–8. Confavreux C, Vukusic S. Natural history of multiple sclerosis: a unifying concept. Brain 2006;129:606–16. Cook DA, Hatala R, Brydges R, Zendejas B, Szostek JH, Wang AT, et al. Technology-enhanced simulation for health professions education: A systematic review and meta-analysis. Journal of the American Medical Association 2011;306:978–88. Daumer M, Neuhaus A, Lederer C, Scholz M, Wolinsky JS, Heiderhoff M. Prognosis of the individual course of disease-steps in developing a decision support tool for multiple sclerosis. BMC Medical Informatics and Decision Making 2007;7:11. Dong Y, Chbat NW, Gupta A, Hadzikadic M, Gajic O. Systems modeling and simulation applications for critical care medicine. Annals of Intensice Care 2012;2:18. Ebers GC. Prognostic factors for multiple sclerosis: the importance of natural history studies. Journal of Neurology 2005;252(Suppl. 3):iii15–20. Edwards A, Elwyn G, Mulley A. Explaining risks: turning numerical data into meaningful pictures. British Medical Journal 2002;324:827–30. Elwyn G, O'Connor A, Stacey D, Volk R, Edwards A, Coulter A, et al. International Patient Decision Aids Standards (IPDAS) Collaboration. Developing a quality criteria framework for patient decision aids: online international Delphi consensus process. British Medical Journal 2006;333:417. Filippini G, Munari L, Incorvaia B, Ebers GC, Polman C, D′Amico R, et al. Interferons in relapsing remitting multiple sclerosis: a systematic review. Lancet 2003;361:545–52. Francis G, Grumser Y, Alteri E, Micaleff A, O'Brien F, Alsop J, et al. Hepatic reactions during treatment of multiple sclerosis with Interferon β-1a, incidence and clinical significance. Drug Safety 2003;26:815–27. Freedman MS, Hughes B, Mikol DD, Bennett R, Cuffel B, Divan V, et al. Efficacy of diasease-modifying therapies in relapsing remitting multiple sclerosis: a systematic comparison. European Neurology 2008;60:1–11. Gray OM, Jolley D, Zwanikken C, Trojano M, Grand'Maison F, Duquette P, et al. The Multiple Sclerosis Severity Score (MSSS) re-examined: EDSS rank stability in the MSBase dataset increases 5 years after onset of multiple sclerosis. Neurology Asia 2008;13:217–9. Heesen C, Solari A, Giordano A, Kasper J, Köpke S. Decisions on multiple sclerosis immunotherapy: new treatment complexities urge patient engagement. Journal of the Neurological Sciences 2011;306(1–2):192–7. Hughes TE, Larson LN. Patient involvement in health care. A procedural justice viewpoint. Medical Care 1991;29:297–303. Jacobs LD, Cookfair DL, Rudick RA, Herndon RM, Richert JR, Salazar AM, et al. The Multiple Sclerosis Collaborative Research Group (MSCRG). Intramuscular interferon beta-1a for disease progression in relapsing multiple sclerosis. Annals of Neurology 1996;39:285–94. Jadad AR, Haynes RB, Hunt D, Browman GP. The internet and evidence-based decision-making: a needed synergy for efficient knowledge management in health care. Canadian Medical Association Journal 2000;162:362–5. Jeffery D, Bashir K, Buchwald L, Coyle P, Freedman M, Markowitz C, et al. Optimizing immunomodulatory therapy for MS patients: an integrated management model. Journal of the Neurological Sciences 2002;201:89–90. Johnson FR, Özdemir S, Hauber AB, Kauf TL. Women's stated willingness to accept perceived risk for vasomotor symptom relief. Journal of Women's Health 2007a;16:1028–40. Johnson FR, Özdemir S, Mansfield CA, Hass S, Miller DW, Siegel CA, et al. Crohn's disease patients' benefit-risk preferences: serious
383
adverse event risks versus treatment efficacy. Gastroenterology 2007;133:769–79. Johnson KP, Brooks BR, Cohen JA, Ford CC, Goldstein J, Lisak RP, et al. The Copolymer 1 Multiple Sclerosis Study Group. Copolymer 1 reduces relapse rate and improves disability in relapsingremitting multiple sclerosis: results of a phase III multicenter, double-blind placebo-controlled trial. Neurology 1995;45: 1268–76. Kaplan B, Brennan PF. Consumer informatics supporting patients as co-producers of quality. Journal of the American Medical Informatics Association 2001;8:209–16. Kremenchutzky M, Rice GPA, Baskerville J, Wingerchuk M, Ebers GC. The natural history of multiple sclerosis: a geographically based study. 9: observations on the progressive phase of rhe disease. Brain 2006;129:584–94. Kuljis J, Paul RJ, Stergioulas LK. Can health care benefit from modeling and simulation methods in the same way as business and manufacturing has? In: Winter simulation conference proceedings 2007; p. 1449–53. Law A, Kelton WD. Simulation modeling and analysis.New York: McGrawHill Science/Engineering/Math; 1999. Levic ZM, Dujmovic I, Pekmezovic T, Jareninski M, Marinkovic J, Stojsavljevic N, et al. Prognostic factors for survival in multiple sclerosis. Multiple Sclerosis 1999:171–8. Mazur DJ, Hickam DH. Patients' interpretations of probability terms. Journal of General Internal Medicine 1991;6:237–40. Miller JH, Page SE. Complex adaptive systems: an introduction to computational models of social life (Princeton studies in complexity).Princeton, New Jersey: Princeton University Press; 2007. O'Connor AM, Wennberg JE, Legare F, Llewellyn-Thomas HA, Moulton BW, Sepucha KR, et al. Toward the ‘tipping point’: decision aids and informed patient choice. Health Affairs 2007;26:716–25. Pachner AR, Steiner I. The multiple sclerosis severity score (MSSS) predicts disease severity over time. Journal of the Neurological Sciences 2009;278:66–70. Polman CH, O'Connor PW, Havrdova E, Hutchinson M, Kappos L, Miller DH, et al. A randomized, placebo-controlled trial of natalizumab for relapsing multiple sclerosis. New England Journal of Medicine 2006;354:899–910. Roxburgh RH, Seaman SR, Masterman T, Hensiek AE, Sawcer SJ, Vukusic S, et al. Multiple Sclerosis Severity Score Using disability and disease duration to rate disease severity. Neurology 2005;64:1144–51. Runmarker B, Andersen O. Prognostic factors in a multiple sclerosis incidence cohort with twenty five years of follow-up. Brain 1993;116:117–34. Sackett DL, Rosenberg WMC, Gray JAM, Haynes RB, Richardson WS. Evidence-based medicine: what it is and what it isn′t. British Medical Journal 1996;312:71–2. Scalfari A, Neuhaus A, Muraro PA, Daumer M, DeLuca GC, Muraro PA, et al. Early relapses, onset of progression, and late outcome in multiple sclerosis. Archives of Neurology 2012. Publisehed online November 19. Sepucha KR, Fowler Jr. FJ, Mulley Jr. AG. Policy support for patientcentered care: the need for measurable improvements in decision quality. Health Affairs (Millwood) 2004(Suppl. Variation):VAR54–62. Small SD. Simulation applications for human factors and systems evaluation. Anesthesiology Clinics 2007;25:237–59. Vukusic S, Confavreux C. Natural history of multiple sclerosis: risk factors and prognostic indicators. Current Opinion in Neurology 2007;20:269–74. Weinshenker BG, Bass B, Rice GPA, Noseworthy JH, Carriere W, Basekerville J, et al. The natural history of multiple sclerosis: a geographically based study: 2. Predictive value of the early clinical course. Brain 1989;112:133–46.
384 Weinshenker BG, Rice GPA, Noseworthy JH, Carriere W, Basekerville J, Ebers GC. The natural history of multiple sclerosis: a geographically based study: 4. Applications to planning and interpretation of clinical therapeutic trials. Brain 1991;114: 1057–67. Wennberg JE, Fisher ES, Skinner JS. Geography and the debate over Medicare reform. Health Affairs (Millwood) 2002(Suppl. Web Exclusives):W96–114. Wilensky U, NetLogo. Evanston IL: Center for connected learning and computer based modeling, Northwestern University; 1999. Avaibable from:.〈http://ccl.northwestern.edu/netlogo/〉. Wolfson C, Confavreux C. A improvements to a simple Markov model of the natural history of multiple sclerosis: I. Short term prognosis. Neuroepidemiology 1987;6:101–15.
M. Veloso Wolfson C, Confavreux CA. Markov model of the natural history of multiple sclerosis. Neuroepidemiology 1985;4:227–39. Woloshin S, Schwartz LM, Moncur M, Gabriel S, Tosteson AN. Assessing values for health: numeracy matters. Medical Decision Making 2001;21:382–90. Wyatt JC, Altman DG. Commentary prognostic models: clinically useful or quickly forgotten. British Medical Journal 1995; 311:1539. Yousry TA, Habil DM, Major EO, Ryschkewitsch C, Fahle G, Fischer S, et al. Evaluation of patients treated with natalizumab for progressive multifocal leukoencephalopathy. New England Journal of Medicine 2006;354:924–33.
Available online at www.sciencedirect.com
journal homepage: www.elsevier.com/locate/msard
An agent-based simulation model for informed shared decision making in multiple sclerosis Mário Veloson,1 Hospital Egas Moniz (Centro Hospitalar de Lisboa Ocidental, EPE), Rua da Junqueira, 126, 1349-019 Lisboa, Portugal Received 11 February 2013; received in revised form 25 March 2013; accepted 7 April 2013
KEYWORDS
Abstract
Multiple sclerosis; Shared decision making; Simulation model; Prognosis; Decision aids; Agent-based system
Shared decision making (SDM) is concerned with patient involvement into medical decisions and chronic conditions such as Multiple sclerosis (MS), with only partially effective treatments leading to potential severe side effects, conflicting evidence, and uncertain evidence on outcomes, constitute a typical condition for SDM. As treatment options increase and patients participate more intensively in decisions, the need for evidence-based information (EBI) becomes clear. Natural history (NH) studies of MS represent the basic sources for required EBI and are especially useful to contribute to the practical exercise of prognosis formulation and to enable the evaluation of effectiveness in the context of treatment. Several of these identify early clinical factors predictive of the course of MS but there is no consensus method for determining the long term progression of disability and evolution of individual patients on the basis of observations on the early stages of the disease, which constitutes a major challenge for the practicing neurologist. Aiming at delivering more reliable prognosis estimation, this study combines the distribution of patients reaching specific levels of disability within defined time periods as determined in NH studies, with disability curves and severity scores as a function of time, in terms of percentiles and deciles respectively, derived from longitudinal data analysis studies. A computer agent-based simulation model was implemented as a comprehensive and easy to utilize tool able to predict and monitor progression of disability in MS patients, and to support the neurologist discussing prognosis scenarios with the individual patient for effective SDM. & 2013 Elsevier B.V. All rights reserved.
1.
n
Tel.: +351 964020950 (Mob); fax: +351 217993913. E-mail addresses: [email protected], [email protected]. 1 Present address: ARN—Anestesia, Reanimação e Neurologia, Lda, Campo Grande 14-6A, 1700-092 Lisboa, Portugal. 2211-0348/$ - see front matter & 2013 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.msard.2013.04.001
Introduction
The goal of both diagnosis and therapeutic decisions is to improve the prognosis for the patient, since prognosis refers to all medical outcomes that may occur during the patient's disease process. This also implies that patient management should be driven by two major prognosis related topics: the natural history (NH), which is the prognosis of the disease without medical interventions, and the prognosis changes as
378 resulting from medical interventions. Additionally, prognosis is a major concern to the patient who wants to be informed about his prospects and prognosis related information is a mandatory requirement for an informed and active participation of the patient on self-clinical decision-making. A major challenge in multiple sclerosis (MS) for the practicing neurologist is to make a prediction of the long term evolution of individual patients on the basis of observations on the early stages of the disease. An immediate effect of such difficulty is to inhibit the communication to the patient of a realistic estimation of his/her evolution, particularly in a long term basis. Prognosis' estimates or prediction can be made in several ways. As opposed to an informal way (e.g. intuitively or using expert opinions), modern patient management requires that appropriateness of medical interventions is supported by scientific evidence, integrating clinical expertise, patient values, and the best research evidence into the decision making process for patient care, the basis for evidencebased medicine (Sackett et al., 1996). Including patient values and preferences in the clinical decision-making is an ethical issue (Hughes and Larson, 1991) and can contribute to improve patient care (O′Connor et al., 2007; Sepucha et al., 2004). Shared decision-making (SDM) is a major component of such patient-centered care, and has been defined as a process that “allows both physicians and patients to honor the values and preferences of the patient, while also permitting the physician to provide medical expertise to promote the patient′s health” (O′Connor et al., 2007). Hence, SDM takes informed consent a step further in the process of communication between a patient and a physician. It is much more than obtaining the patient′s authorization or agreement to undergo a specific medical intervention; SDM means to involve the patient actively on the medical decisions and such patient empowerment also means giving the patient more responsibility. Chronic conditions such as MS, with only partially effective treatments with potential severe side effects, conflicting evidence, and uncertain evidence on outcomes, where the benefit–harm ratio is short and/or doubtful, or when available options have different benefit–harm profiles that patients value differently, constitute a typical condition for this SDM and enforces its need (Heesen et al., 2011; Wennberg et al., 2002). As treatment options increase and patients participate more intensively in decisions, the need for evidence-based information becomes clear. There is an evidence that patients may, in exchange for therapeutic benefits, be willing to accept greater levels of risk than are actually posed by some therapies (Calfee, 2006; Johnson et al., 2007a, 2007b). Drug therapies for MS offer a range of potential benefits, but they may also involve lifethreatening risks, including liver failure, leukemia, and progressive multifocal leukoencephalopathy (Brassat et al., 2002; Francis et al., 2003; Yousry et al., 2006). Yet patients may be misinformed particularly by means of internet information, misinterpret the results of scientific research (Jadad et al., 2000; Kaplan and Brennan, 2001), have unrealistic expectations of treatment benefits and harms, and clinicians may be poor judges of a patient′s values (O′Connor et al., 2007). There also is evidence that patient decision aids are better than usual care in improving patients' knowledge and
M. Veloso expectations about interventions, as well as improving agreement between values and choice (Elwyn et al., 2006). A diversity of methods has emerged for decision support from different scientific fields such as Statistics, Decision Analysis and Artificial Intelligence. Cox models, recursive partitioning analysis, Weibull models, decision trees, Markov models, Partially Observable Markov Decision Process, Bayesian networks and influence diagrams. No matter the approach, the first step is to build up a prognostic model able to predict the probability of some outcome as optimally as possible. Several studies have been performed to identify early clinical factors predictive of the MS course (Levic et al., 1999; Runmarker and Andersen, 1993; Weinshenker et al., 1989, 1991; Kremenchutzky et al., 2006; Confavreux and Vukusic, 2006; Ebers, 2005; Vukusic and Confavreux, 2007; Scalfari et al., 2012), such as: gender, disease course, age at onset of disease, initial symptoms, number of functional systems involved, first interval attack, attack frequency, and incomplete remission after the first episode. However, the majority of work in this area is not focused on the individual prognosis to the patient; neither does address treatment effects on the NH of the disease. Daumer et al. (2007), describe an online analytical processing tool that matches the characteristics of a given patient with the most similar patients of the Sylvia Lawry Centre for Multiple Sclerosis Research database. An “individual risk profile” in terms of the disease course of all similar patients in the database is displayed, hence enabling to project a hypothesized outcome for that patient (Daumer et al., 2007). The main limitations of this tool are related to the characteristics of the patients included in the database. The clinical data are derived only from the placebo groups of randomized clinical trials, and the respective observation period is limited to a maximum of three years (Daumer et al., 2007). In the study of Wolfson and Confavreux (1985, 1987), a Markov model is proposed to represent the disease course by means of transitions between disease states, as to evaluate the effect of prognostic factors on those transitions. Because Markov processes are memory less, once a state is known, the future evolution of the disease is independent of the past evolution. This limitation is handled in the work of Bergamaschi et al. (2007, 2001), by proposing a Bayesian model specifying the full joint probability distribution for a set of random variables that characterize the entire course of the disease. The risk of reaching secondary progression was significantly related to specific clinical factors presented during the first year of the disease, all of them associated with a specific statistical weight, the Bayesian local relative risk, used to calculate the Bayesian Risk Estimate for MS (BREMS) score for any given patient (Bergamaschi et al., 2007, 2001). However, no other prognostic outcomes are provided. Achiron et al. (2003) use NH information from a large database to generate longitudinal disability curves for prediction of disease progression based on the mean Expanded Disability Status Scale (EDSS) scores from the first year of disease onset represented as a major percentile group. These curves represent cohort percentiles and enable to foresee the relative risk of disease progression, as well as to identify deviations in the curves (Achiron
An agent-based simulation model for informed shared decision making in multiple sclerosis et al., 2003). Similarly, the global Multiple Sclerosis Severity Score (MSSS) (Roxburgh et al., 2005) has been proposed as a population disability assessment tool enabling comparisons of relative disease severity at all EDSS levels for a given disease duration. A more recent study (Gray et al., 2008) confirmed the validity of the MSSS as a 5-year severity rank predictor in individual patients. In spite of the increasing number of prognostic models developed in Medicine, the fact is that such prognostic models are seldom used in the clinical practice (Wyatt and Altman, 1995; Dong et al., 2012). A major challenge, often underestimated, consists of the ability to provide the relevant information for decision-making in a comprehensive
Fig. 1
379
and easy to process format both for patients and physicians. Different patients have different information needs, and different ways of decision-making (Mazur and Hickam, 1991). Moreover, many people, including physicians, have difficulty in process critically quantitative information (Woloshin et al., 2001). Considering that most information relevant to decision-making is inherently probabilistic, new ways for presenting such information are required (Edwards et al., 2002; Heesen et al., 2011). Building and using models is part of everybody′s life, being simplification and abstraction of the real system the key in modeling. Agent-based modeling (ABM) is a type of modeling in which the focus is on representing agents (such
Simulation model user interface.
380
M. Veloso
as people) and their interactions (Miller and Page, 2007), enabling to effectively capture a very rich set of complex behaviors and interactions, hence highly suited to modeling complex phenomena. ABM capabilities explain its extensive use in a diversity of fields, including decision-support and Medicine (Cook et al., 2011; Small, 2007). The main goal of the present work is to provide the clinician with an easy to use simulation tool concerning individual long term (30 years) disability prediction and treatment effect on the estimated individual prognosis, in order to enable a comprehensive interaction with the patient with MS for effective SDM.
2. 2.1.
Methods The simulation model
The simulation model was implemented using the Netlogo 5.0.2 programmable modeling environment (Wilensky and NetLogo, 1999). Netlogo is a programming language with predominantly agent-oriented attributes, with unique capabilities that make it extremely powerful for producing and visualizing simulations of multi-agents systems. Taking into account that available disease-modifying agents are only indicated for the relapsing-remitting form of MS, the proposed simulation model only addresses this form of disease. For simulation purposes the model uses virtual population of 100 agents (simulating persons with relapsing-remitting MS) and all calculations are reflected in this population by means of different sorts of graphics (Fig. 1). Basically, disease progression is derived from the number of years from disease onset to EDSS scores 4, 6 and 7, according to published data of the Lyon database (Confavreux and Compston, 2006; Confavreux et al., 2003, 2000; Confavreux and Vukusic, 2006). The 95% confidence interval of the time to assignment of the different EDSS scores was used as reference for building the state transition rules that guarantee a proper probability distribution of the 100 agents (virtual MS patients) along the time. Transitions intervals are anticipated or delayed in accordance to the individual patient characteristics as entered by the user. These characteristics are the clinical variables used to build up the BREMS score (Bergamaschi et al., 2007, 2001), corresponding to known prognostic factors: age at onset of disease, gender, sphincter onset, pure motor onset, motorsensory onset, sequel after onset, number of involved functional systems at onset, number of sphincter plus motor relapses, EDSS≥4. Dependent on selected patient characteristics, the system will display a 30 years prediction of the probabilities to be at EDSS scores o4, 4, 6 and 7 in two graphics of a group of 100 patients with similar clinical characteristics. One graphic, will display 100 “persons” with different color intensity (pink for females, blue for males) representing the different EDSS possibilities. The second graphic, will present the yearly distribution of the different EDSS states along the 30 years projection. The chance of being in a progressive phase of the disease at 10 years after onset, and the value of patient quality adjusted life years (QALYs) after 30 years will be also displayed numerically in two distinct
boxes. Patient QALYs are calculated as a function of the number of years at the different EDSS states estimated for the individual patient. By providing the system with the EDSS in subsequent years, the prognosis estimation in terms of disability progression will be refined. The patient EDSS score will be graphically situated in relation with the 10 years disability progression curves representing the respective annually percentile figures (Achiron et al., 2003). On the other hand, the yearly introduced EDSSs will enable to calculate the MSSS for those years, hence to make a 30 years projection corresponding to the maximum EDSS score expected to the derived decile (Roxburgh et al., 2005). Besides the predictive facilities, these graphics will enable the user to monitor disability progression along the years, hence to early recognize any up-deviation which certainly correspond to a clinical deterioration. The simulation model enables three ways for treatment assessment. Firstly, pressing the “assess” button after selecting a treatment from the list, a new 30 years prediction is made by the system reflecting the treatment effect in terms of “Absolute Risk Reduction” on two variables in accordance to published data in the literature: relapse free and progression free at 2 years treatment (Filippini et al., 2003; Freedman et al., 2008; Jacobs et al., 1996; Johnson et al., 1995; Polman et al., 2006). Treatment effect is calculated through the impact of those two factors upon the time to reach the different EDSS states, and this will be reflected in the estimation of the EDSS distribution along the 30 years that will be displayed, together a prediction of the QALYs gain at 30 years, Table 1 Validation group as compared to overall patient characteristics.
Gender Male—N (%) Female—N (%)
RRMSa (N =127)
(N =50)
PPMSmb (N =9)
31 (24.4%) 96 (75.6%)
13 (26%) 37 (74%)
5 (55.6%) 4 (44.4%)
Age at disease onset Mean (SDc) 28.0 (7 17.4) Median 27.8
28.42 (7 8.39) 25.65
Functional system affected—N (%) Motor 33 (26.0%) 12 (24.0%) Sensory 37 (29.1%) 18 (36.0%) Coordenation 7 (5.5%) 1 (2.0%) Visual 34 (26.8%) 14 (28.0%) Brainstem 11 (8.7%) 5 (10.0%) Undefined 5 (3.9%) 0 Interval 1st–2nd relapse 4.8 (711.1) 4.66 Mean (SDc) (76.03) Median 2.0 2.59 a
RRMS: relapsing-remitting multiple sclerosis. PPMS: primary progressive multiple sclerosis. c SD: standard deviation. b
An agent-based simulation model for informed shared decision making in multiple sclerosis estimated as described before. Secondly, treatment effect can be monitored through the early recognition of any deviation in the disability curves and MSSS progression graphics. Finally, by introducing yearly data in addition to the EDDS (number of mild and severe relapses, and number of new Magnetic Ressonance Imaging lesions), a textual recommendation is displayed in accordance to published consensus adopted criteria (Jeffery et al., 2002).
2.2.
Model validation
Data for the validation of the simulation model was retrospectively obtained from the hospital medical records of a group of 50 patients selected from the total of the 173 patients observed by the author between January and July of 2012 in the MS outpatient clinic of Hospital Egas Moniz. The following criteria were used for selecting the validation group: Patients with relapsing-remitting multiple sclerosis (RRMS) with at least 10 years of disease evolution and sufficient data in their medical records, namely the number of relapses at years 1, 2, 5 and 10, and the EDSS scores at years 2, 5 and 10. The study has been approved by the Hospital Ethics Committee and performed in accordance with the ethical standards laid down in the 1964 Declaration of Helsinki and its later amendments. The validation procedure consisted of assessing the concordance degree between the simulation model predictions and the real value presented by the patients, for 10 and 20 years estimations, using patient EDSS at years 2 and 5 of disease evolution.
3.
Results
Thirty seven patients among the 173 patients observed were discharged because of lack of data in their clinical record together with patient incapacity in describing the facts related to their initial clinical history. Data from the validation group of 50 patients is presented in Table 1, as compared to the data of the overall 136 patients. All the data and validation procedures described next focus on this subgroup of 50 patients. The mean age at onset was 28.4 years (7 8.4, median 25,6), mean disease evolution 17.1 (77.8), and the mean EDSS at the last observation was 3 (7 2.2), median 2. The number of relapses and EDSS scores were recorded (Table 2); the mean time to reach EDSS 4 was 12.4 (76.9) years and EDSS 6 was 20.9 (78.8), median 11 and 18 respectively.
Table 2
b
Table 3 presents the correlation between the number of relapses and EDSS at years 5 and 10, while Table 4 presents the correlation between EDSS scores at different years. As to the model 10 years prediction, a strong correlation was found both with the disability curves (Pearson coefficient= 0.75; po0.0001) and the MSSS projection (Pearson coefficient = 0.67; po0.0001). If one considers the minimum–maximum interval of MSSS estimation, in 76% of cases the system made an adequate prediction. As to the 20 years prediction of the MSSS, similar correlation degree (Pearson coefficient = 0.67; p= 0.0016) and probability of min–max correct estimation (76%—13 of 17 patients) were found. Because lack of data it was not possible to evaluate the model behavior concerning the patient characteristics at disease onset, neither treatment effect.
4.
Discussion
NH studies of MS are especially useful to contribute to the practical exercise of prognosis formulation and to enable the evaluation of effectiveness in the context of treatment. Analysis of the MS' NH using different statistical approaches, proposed similar conclusions concerning the prognosis influence of some basic predictors. However, because of high variation amongst patients' disease course only rough predictions for the individual patient are possible. Table 3 Correlation between number of relapses and EDSS.
Number of relapses at year 2 Number of relapses at year 5 a
EDSSa at year 5
EDSSa at year 10
0.50 (p =0.0001) 0.60 (po0.0001)
0.26 (p =0.031) 0.40 (p =0.002)
EDSS: Expanded Disability Status Scale.
Table 4 Correlation between EDSSa at early stages with EDSSa at 10 years.
EDSSa at year 2 EDSSa at year 5 a
EDSSa at year 5
EDSSa at year 10
0.75 (po0.0001) –
0.60 (po0.0001) 0.77 (po0.0001)
EDSS: Expanded Disability Status Scale.
Recorded relapses and EDSSa scores.
Mean number of relapses (SDb) EDSS mean score (SDb) a
381
EDSS: Expanded Disability Status Scale. SD: standard deviation.
Year 1
Year 2
Year 5
Year 10
0.47 (70,7) –
0.80 (71.1) 0.76 (71.0)
1.60 (71.4) 1.27 (71.0)
2,76 (72.0) 2,02 (72.0)
382 The present study combined the results of longitudinal studies (Achiron et al., 2003; Bergamaschi et al., 2007, 2001; Roxburgh et al., 2005) with previous NH studies and disease prognostic factors, in order to provide clinicians with an easy to use tool concerning individual long term disability prediction and treatment effect on the estimated individual prognosis, enabling a comprehensive interaction with the patient with MS for effective SDM. For such purpose, a computer model able to simulate individual patient prognosis was implemented. Based on patient characteristics at disease onset, the model provides individual 30 years disability prediction derived from the time to reach EDSS scores 4, 6 and 7. It also enables to asses treatment effect on the estimated individual prognosis, as well as to monitor deviations in the disability curves (percentiles) and in the MSSS (deciles), as patient data in subsequent years is provided. Modeling and simulation have been used in a variety of scientific domains (Law and Kelton, 1999), and significant improvement in decision-making, efficiency and quality reported (Kuljis et al., 2007). The main advantage of modeling is to facilitate to understand the behavior of a real system and then to test it through a variety of simulations of different scenarios. The simulation model now presented is drawn upon available scientific evidence relevant for the defined purpose. However, in general, available data are scarce and/or not adequate for the modeling requirements limiting model validation. Ultimately, the main limitations of the current model are the limitations of the data that the model proposes to represent and process. In order to ensure consistency, a single data source was used but a metaanalysis of several sources is also a valid option. Promising results were obtained through performed simulations with data from real patients when complementing model estimations based on patient characteristics at disease onset with initial defined percentiles and deciles. Using the initial EDSS it is already possible to predict progression along percentiles and deciles regardless of relapses, since relapses seem not significantly influence long term disability (Confavreux et al., 2000), as also documented in the present study. Initial estimations can be later refined by specifying EDSS scores in subsequent years. A major added value of the current model consists of its monitoring capabilities particularly by enabling the clinician to be aware early of any change to expected course of disease. Any up-deviation from assigned percentile or decile means unexpected disease deterioration, hence the need for intervention (e.g. treatment switch or escalation). This capability is further refined by means of the criteria for the identification of sub-optimal treatment response. The results obtained through the assessment of the different disease-modifying agents effect in terms of QALYs gain after 30 years deserve a comment. Whether the reduced gain obtained is a simulation problem or it represents actually a low impact of treatment into long term accumulated quality of life, remains a question to be solved. Nevertheless, lack of a major impact of disease-modifying agents upon disease severity as measured by the MSSS has been previously reported (Pachner and Steiner, 2009). Previewed future model developments are twofold. On one hand, improvement of the data related to NH of the
M. Veloso disease supporting the model is an ongoing process. In particular, partnership with other research and/or MS centers is being envisaged. The overall idea is to validate the model with different data sets and to evaluate model simulations as to the impact of patient characteristics (e.g. prognostic factors) at disease onset upon the time to reach the different EDSS states. Similarly, treatment effect on state transition intervals also needs deeper testing. On the other hand, model integration in a patient record system seems to be a natural extension to current achievements. NetLogo, besides enabling to delivering models as java applets, hence easily incorporated into web pages, also provides a programming interface to other programming languages.
5.
Conclusions
To be useful for clinicians, a MS prognosis model should provide a reliable prediction at the individual level and a way to monitor disability progression in MS patients. By combining two different approaches of modeling disability progression, the proposed model of prognosis prediction seems to be at least as good as the ones that can be extracted from available scientific evidence. Therefore, one can conclude that the current model is a valuable tool and, because of its conceptual simplicity and easy utilization, this type of simulation model can be used in everyday clinical practice both to monitor progression of disability, and to support the neurologist discussing prognosis scenarios with the individual patient for effective SDM.
Conflict of interest The author has no conflicts of interest or disclosure to report. This work has received no financial support.
References Achiron A, Barak Y, Rotstein Z. Longitudinal disability curves for predicting the course of relapsing-remitting multiple sclerosis. Multiple Sclerosis 2003;9:486–91. Bergamaschi R, Quaglini S, Trojano M, Amato MP, Tavazzi E, Paolicelli D, et al. Early prediction of the long term evolution of multiple sclerosis: the Bayesian Risk Estimate for Multiple Sclerosis (BREMS) score. Journal of Neurology, Neurosurgery, and Psychiatry 2007;78:757–9. Bergamaschi R, Berzuini C, Romani A, Cosi V. Predicting secondary progression in relapsing-remitting multiple sclerosis: a Bayesian analysis. Journal of the Neurological Sciences 2001;189:13–21. Brassat D, Recher C, Waubant E, Le Page E, Rigal-Huguet F, Laurent G, et al. Therapy-related acute myeloblastic leukemia after mitoxantrone treatment in a patient with MS. Neurology 2002;59:954–5. Calfee, J.E. A representative survey of MS patients on attitudes toward the benefits and risks of drug therapy. AEI-Brookings Joint Center for Regulatory Studies; 2006. Confavreux C, Compston A. The natural history of multiple sclerosis. In: Compston A, editor. McAlpine's multiple sclerosis. 4th ed. London: Churchill Livingstone Elsevier; 2006. p. 183–272. Confavreux C, Vukusic S, Adeleine P. Early clinical predictors and progression of irreversible disability in multiple sclerosis: an amnesic process. Brain 2003;126:770–82.
An agent-based simulation model for informed shared decision making in multiple sclerosis Confavreux C, Vukusic S, Moreau T, Adeleine P. Relapses and progression of disability in multiple sclerosis. New England Journal of Medicine 2000;343:1430–8. Confavreux C, Vukusic S. Natural history of multiple sclerosis: a unifying concept. Brain 2006;129:606–16. Cook DA, Hatala R, Brydges R, Zendejas B, Szostek JH, Wang AT, et al. Technology-enhanced simulation for health professions education: A systematic review and meta-analysis. Journal of the American Medical Association 2011;306:978–88. Daumer M, Neuhaus A, Lederer C, Scholz M, Wolinsky JS, Heiderhoff M. Prognosis of the individual course of disease-steps in developing a decision support tool for multiple sclerosis. BMC Medical Informatics and Decision Making 2007;7:11. Dong Y, Chbat NW, Gupta A, Hadzikadic M, Gajic O. Systems modeling and simulation applications for critical care medicine. Annals of Intensice Care 2012;2:18. Ebers GC. Prognostic factors for multiple sclerosis: the importance of natural history studies. Journal of Neurology 2005;252(Suppl. 3):iii15–20. Edwards A, Elwyn G, Mulley A. Explaining risks: turning numerical data into meaningful pictures. British Medical Journal 2002;324:827–30. Elwyn G, O'Connor A, Stacey D, Volk R, Edwards A, Coulter A, et al. International Patient Decision Aids Standards (IPDAS) Collaboration. Developing a quality criteria framework for patient decision aids: online international Delphi consensus process. British Medical Journal 2006;333:417. Filippini G, Munari L, Incorvaia B, Ebers GC, Polman C, D′Amico R, et al. Interferons in relapsing remitting multiple sclerosis: a systematic review. Lancet 2003;361:545–52. Francis G, Grumser Y, Alteri E, Micaleff A, O'Brien F, Alsop J, et al. Hepatic reactions during treatment of multiple sclerosis with Interferon β-1a, incidence and clinical significance. Drug Safety 2003;26:815–27. Freedman MS, Hughes B, Mikol DD, Bennett R, Cuffel B, Divan V, et al. Efficacy of diasease-modifying therapies in relapsing remitting multiple sclerosis: a systematic comparison. European Neurology 2008;60:1–11. Gray OM, Jolley D, Zwanikken C, Trojano M, Grand'Maison F, Duquette P, et al. The Multiple Sclerosis Severity Score (MSSS) re-examined: EDSS rank stability in the MSBase dataset increases 5 years after onset of multiple sclerosis. Neurology Asia 2008;13:217–9. Heesen C, Solari A, Giordano A, Kasper J, Köpke S. Decisions on multiple sclerosis immunotherapy: new treatment complexities urge patient engagement. Journal of the Neurological Sciences 2011;306(1–2):192–7. Hughes TE, Larson LN. Patient involvement in health care. A procedural justice viewpoint. Medical Care 1991;29:297–303. Jacobs LD, Cookfair DL, Rudick RA, Herndon RM, Richert JR, Salazar AM, et al. The Multiple Sclerosis Collaborative Research Group (MSCRG). Intramuscular interferon beta-1a for disease progression in relapsing multiple sclerosis. Annals of Neurology 1996;39:285–94. Jadad AR, Haynes RB, Hunt D, Browman GP. The internet and evidence-based decision-making: a needed synergy for efficient knowledge management in health care. Canadian Medical Association Journal 2000;162:362–5. Jeffery D, Bashir K, Buchwald L, Coyle P, Freedman M, Markowitz C, et al. Optimizing immunomodulatory therapy for MS patients: an integrated management model. Journal of the Neurological Sciences 2002;201:89–90. Johnson FR, Özdemir S, Hauber AB, Kauf TL. Women's stated willingness to accept perceived risk for vasomotor symptom relief. Journal of Women's Health 2007a;16:1028–40. Johnson FR, Özdemir S, Mansfield CA, Hass S, Miller DW, Siegel CA, et al. Crohn's disease patients' benefit-risk preferences: serious
383
adverse event risks versus treatment efficacy. Gastroenterology 2007;133:769–79. Johnson KP, Brooks BR, Cohen JA, Ford CC, Goldstein J, Lisak RP, et al. The Copolymer 1 Multiple Sclerosis Study Group. Copolymer 1 reduces relapse rate and improves disability in relapsingremitting multiple sclerosis: results of a phase III multicenter, double-blind placebo-controlled trial. Neurology 1995;45: 1268–76. Kaplan B, Brennan PF. Consumer informatics supporting patients as co-producers of quality. Journal of the American Medical Informatics Association 2001;8:209–16. Kremenchutzky M, Rice GPA, Baskerville J, Wingerchuk M, Ebers GC. The natural history of multiple sclerosis: a geographically based study. 9: observations on the progressive phase of rhe disease. Brain 2006;129:584–94. Kuljis J, Paul RJ, Stergioulas LK. Can health care benefit from modeling and simulation methods in the same way as business and manufacturing has? In: Winter simulation conference proceedings 2007; p. 1449–53. Law A, Kelton WD. Simulation modeling and analysis.New York: McGrawHill Science/Engineering/Math; 1999. Levic ZM, Dujmovic I, Pekmezovic T, Jareninski M, Marinkovic J, Stojsavljevic N, et al. Prognostic factors for survival in multiple sclerosis. Multiple Sclerosis 1999:171–8. Mazur DJ, Hickam DH. Patients' interpretations of probability terms. Journal of General Internal Medicine 1991;6:237–40. Miller JH, Page SE. Complex adaptive systems: an introduction to computational models of social life (Princeton studies in complexity).Princeton, New Jersey: Princeton University Press; 2007. O'Connor AM, Wennberg JE, Legare F, Llewellyn-Thomas HA, Moulton BW, Sepucha KR, et al. Toward the ‘tipping point’: decision aids and informed patient choice. Health Affairs 2007;26:716–25. Pachner AR, Steiner I. The multiple sclerosis severity score (MSSS) predicts disease severity over time. Journal of the Neurological Sciences 2009;278:66–70. Polman CH, O'Connor PW, Havrdova E, Hutchinson M, Kappos L, Miller DH, et al. A randomized, placebo-controlled trial of natalizumab for relapsing multiple sclerosis. New England Journal of Medicine 2006;354:899–910. Roxburgh RH, Seaman SR, Masterman T, Hensiek AE, Sawcer SJ, Vukusic S, et al. Multiple Sclerosis Severity Score Using disability and disease duration to rate disease severity. Neurology 2005;64:1144–51. Runmarker B, Andersen O. Prognostic factors in a multiple sclerosis incidence cohort with twenty five years of follow-up. Brain 1993;116:117–34. Sackett DL, Rosenberg WMC, Gray JAM, Haynes RB, Richardson WS. Evidence-based medicine: what it is and what it isn′t. British Medical Journal 1996;312:71–2. Scalfari A, Neuhaus A, Muraro PA, Daumer M, DeLuca GC, Muraro PA, et al. Early relapses, onset of progression, and late outcome in multiple sclerosis. Archives of Neurology 2012. Publisehed online November 19. Sepucha KR, Fowler Jr. FJ, Mulley Jr. AG. Policy support for patientcentered care: the need for measurable improvements in decision quality. Health Affairs (Millwood) 2004(Suppl. Variation):VAR54–62. Small SD. Simulation applications for human factors and systems evaluation. Anesthesiology Clinics 2007;25:237–59. Vukusic S, Confavreux C. Natural history of multiple sclerosis: risk factors and prognostic indicators. Current Opinion in Neurology 2007;20:269–74. Weinshenker BG, Bass B, Rice GPA, Noseworthy JH, Carriere W, Basekerville J, et al. The natural history of multiple sclerosis: a geographically based study: 2. Predictive value of the early clinical course. Brain 1989;112:133–46.
384 Weinshenker BG, Rice GPA, Noseworthy JH, Carriere W, Basekerville J, Ebers GC. The natural history of multiple sclerosis: a geographically based study: 4. Applications to planning and interpretation of clinical therapeutic trials. Brain 1991;114: 1057–67. Wennberg JE, Fisher ES, Skinner JS. Geography and the debate over Medicare reform. Health Affairs (Millwood) 2002(Suppl. Web Exclusives):W96–114. Wilensky U, NetLogo. Evanston IL: Center for connected learning and computer based modeling, Northwestern University; 1999. Avaibable from:.〈http://ccl.northwestern.edu/netlogo/〉. Wolfson C, Confavreux C. A improvements to a simple Markov model of the natural history of multiple sclerosis: I. Short term prognosis. Neuroepidemiology 1987;6:101–15.
M. Veloso Wolfson C, Confavreux CA. Markov model of the natural history of multiple sclerosis. Neuroepidemiology 1985;4:227–39. Woloshin S, Schwartz LM, Moncur M, Gabriel S, Tosteson AN. Assessing values for health: numeracy matters. Medical Decision Making 2001;21:382–90. Wyatt JC, Altman DG. Commentary prognostic models: clinically useful or quickly forgotten. British Medical Journal 1995; 311:1539. Yousry TA, Habil DM, Major EO, Ryschkewitsch C, Fahle G, Fischer S, et al. Evaluation of patients treated with natalizumab for progressive multifocal leukoencephalopathy. New England Journal of Medicine 2006;354:924–33.
