No full-text available
To read the full-text of this research,
you can request a copy directly from the author.
MILC: A Microsimulation Model of the Natural History of Lung Cancer
Abstract
The Microsimulation Lung Cancer (MILC) model was developed to simulate individual trajectories and predict outcomes of lung cancer for populations. The model describes the natural history of lung cancer from a disease-free state to death. Predictions of individual trajectories depend on a set of covariates including age, sex, and smoking behaviors. The module presented here is designed as part of a comprehensive decision-making toolkit for evaluating lung cancer prevention, screening and treatment policies. The MILC package implements the model in the open-source statistical software R. This paper introduces the main components, simulation algorithm, and specifics of the MILC model, validates it by reproducing observed lung cancer incidence trends in the US population, and uses it to make plausible predictions for 50-year-old men and women with a range of smoking histories.
Mathematical health policy models, including microsimulation models (MSMs), are widely used to simulate complex processes and predict outcomes consistent with available data. Calibration is a method to estimate parameter values such that model predictions are similar to observed outcomes of interest. Bayesian calibration methods are popular among the available calibration techniques, given their strong theoretical basis and flexibility to incorporate prior beliefs and draw values from the posterior distribution of model parameters and hence the ability to characterize and evaluate parameter uncertainty in the model outcomes. Approximate Bayesian computation (ABC) is an approach to calibrate complex models in which the likelihood is intractable, focusing on measuring the difference between the simulated model predictions and outcomes of interest in observed data. Although ABC methods are increasingly being used, there is limited practical guidance in the medical decision-making literature on approaches to implement ABC to calibrate MSMs. In this tutorial, we describe the Bayesian calibration framework, introduce the ABC approach, and provide step-by-step guidance for implementing an ABC algorithm to calibrate MSMs, using 2 case examples based on a microsimulation model for dementia. We also provide the R code for applying these methods.
Calibration of a microsimulation model (MSM) is a challenging but crucial step for the development of a valid model. Numerous calibration methods for MSMs have been suggested in the literature, most of which are usually adjusted to the specific needs of the model and based on subjective criteria for the selection of optimal parameter values. This article compares 2 general approaches for calibrating MSMs used in medical decision making, a Bayesian and an empirical approach. We use as a tool the MIcrosimulation Lung Cancer (MILC) model, a streamlined, continuous-time, dynamic MSM that describes the natural history of lung cancer and predicts individual trajectories accounting for age, sex, and smoking habits. We apply both methods to calibrate MILC to observed lung cancer incidence rates from the Surveillance, Epidemiology and End Results (SEER) database. We compare the results from the 2 methods in terms of the resulting parameter distributions, model predictions, and efficiency. Although the empirical method proves more practical, producing similar results with smaller computational effort, the Bayesian method resulted in a calibrated model that produced more accurate outputs for rare events and is based on a well-defined theoretical framework for the evaluation and interpretation of the calibration outcomes. A combination of the 2 approaches is an alternative worth considering for calibrating complex predictive models, such as microsimulation models.
The National Lung Screening Trial (NLST) demonstrated that low-dose computed tomography (LDCT) screening reduces lung cancer mortality in a high-risk U.S. population. A microsimulation model of LDCT screening was developed to estimate the impact of introducing population-based screening in Canada.
LDCT screening was simulated using the lung cancer module of the Cancer Risk Management Model (CRMM-LC), which generates large, representative samples of the Canadian population from which a cohort with characteristics similar to NLST participants was selected. Screening parameters were estimated for stage shift, LDCT sensitivity and specificity, lead time, and survival to fit to NLST incidence and mortality results. The estimation process was a step-wise directed search.
Simulated mortality reduction from LDCT screening was 23% in the CRMM-LC, compared with 20% in the NLST. The difference in the number of lung cancer cases over six years varied by, at most, 2.3% in the screen arm. The difference in cumulative incidence at six years was less than 2% in both screen and control arms. The estimated percentage over-diagnosed was 24.8%, which was 6% higher than NLST results.
Simulated screening reproduces NLST results. The CRMM-LC can evaluate a variety of population-based screening strategies. Sensitivity analyses are recommended to provide a range of projections to reflect model uncertainty.
To provide a better understanding of the relationship between primary tumor growth rates and metastatic burden, we present a method that bridges tumor growth dynamics at the population level, extracted from the SEER database, to those at the tissue level. Specifically, with this method, we are able to relate estimates of tumor growth rates and metastatic burden derived from a population-level model to estimates of the primary tumor vascular response and the circulating tumor cell (CTC) fraction derived from a tissue-level model. Variation in the population-level model parameters produces differences in cancer-specific survival and cure fraction. Variation in the tissue-level model parameters produces different primary tumor dynamics that subsequently lead to different growth dynamics of the CTCs. Our method to bridge the population and tissue scales was applied to lung and breast cancer separately, and the results were compared. The population model suggests that lung tumors grow faster and shed a significant number of lethal metastatic cells at small sizes, whereas breast tumors grow slower and do not significantly shed lethal metastatic cells until becoming larger. Although the tissue-level model does not explicitly model the metastatic population, we are able to disengage the direct dependency of the metastatic burden on primary tumor growth by introducing the CTC population as an intermediary and assuming dependency. We calibrate the tissue-level model to produce results consistent with the population model while also revealing a more dynamic relationship between the primary tumor and the CTCs. This leads to exponential tumor growth in lung and power law tumor growth in breast. We conclude that the vascular response of the primary tumor is a major player in the dynamics of both the primary tumor and the CTCs, and is significantly different in breast and lung cancer. Cancer Res; 74(2); 1-10. ©2014 AACR.
A major challenge in the measurement of well-being and progress is to link indicators of high-level societal outcomes with specific policy interventions. This is important not only for better informing the public, but also to provide the means for policy makers and advisors to assess the impacts of their policies and programmes and to increase their effectiveness and cost-efficiency. This paper looks at four major areas of social policies– health status, literacy and learning, economic security, and economic inequality– with the aim of understanding how to link broad outcome measures of progress in these areas, on the one hand, and the policies bearing on them, on the other. Emphasis is given to the powerful benefits to be derived from coupling longitudinal, multivariate data and powerful statistical methods with recently developed analytical tools such as micro-simulation. The paper also emphasises the need for “principled” summary indicators, i.e. indicators embedded within coherent data systems, and the importance of internationally comparable data based on common concepts and definitions.Lorsqu’on mesure le bien-être et le progrès, l’une des principales difficultés consiste à relier les indicateurs de résultats sociaux à des actions spécifiques. Cela est crucial, non seulement pour mieux informer le public, mais aussi pour permettre aux acteurs politiques d’évaluer l’incidence de leurs actions et de leurs programmes, afin de leur permettre d’accroître l’efficacité et l’efficience des mesures en place. Ce rapport porte sur cinq domaines majeurs, pour lesquels le suivi des moteurs du progrès social ou du bien-être est essentiel : i) la santé, ii) l’alphabétisation et l’apprentissage iii) la sécurité économique, iv) les inégalités économiques et v) le manque de temps – l’objectif étant de comprendre comment améliorer les liens entre, d’une part, les indicateurs généraux du progrès dans ces domaines, et d’autre part, les outils permettant d’influer sur les résultats. L’accent est mis sur les grands avantages que présente l’association entre des données longitudinales multi variées, de puissantes méthodes statistiques et des outils d’analyse récents tels que la micro-simulation. Sont également soulignés : la nécessité d’utiliser des indicateurs synthétiques structurés sur des principes définis et s’inscrivant dans des systèmes de données cohérents, et l’importance de disposer de données comparables à l’échelle internationale, fondées sur des définitions et des concepts communs.
This paper presents an overview of microsimulation as a method to evaluate health and health care policies and interventions. After presenting a brief survey of microsimulation models and applications we describe the main features of the approach and how these are implemented in practice. We pay particular attention to the innovative features of dynamic microsimulation as a method of ex-ante policy evaluation. The final section presents a critical overview of the most recent health-dedicated dynamic microsimulation models including POHEM and FEM, two of the most comprehensive dynamic microsimulation models for health. We describe how these models are used to simulate lifecycle health trajectories and associated health care costs under competing policy scenarios to illustrate the power of microsimulation as a valid and relevant tool for policy evaluation.
Background
Computer simulation models are used increasingly to support public health research and policy, but questions about their quality persist. The purpose of this article is to review the principles and methods for validation of population-based disease simulation models.
Methods
We developed a comprehensive framework for validating population-based chronic disease simulation models and used this framework in a review of published model validation guidelines. Based on the review, we formulated a set of recommendations for gathering evidence of model credibility.
Results
Evidence of model credibility derives from examining: 1) the process of model development, 2) the performance of a model, and 3) the quality of decisions based on the model. Many important issues in model validation are insufficiently addressed by current guidelines. These issues include a detailed evaluation of different data sources, graphical representation of models, computer programming, model calibration, between-model comparisons, sensitivity analysis, and predictive validity. The role of external data in model validation depends on the purpose of the model (e.g., decision analysis versus prediction). More research is needed on the methods of comparing the quality of decisions based on different models.
Conclusion
As the role of simulation modeling in population health is increasing and models are becoming more complex, there is a need for further improvements in model validation methodology and common standards for evaluating model credibility.
Microsimulation models that describe disease processes synthesize information from multiple sources and can be used to estimate the effects of screening and treatment on cancer incidence and mortality at a population level. These models are characterized by simulation of individual event histories for an idealized population of interest. Microsimulation models are complex and invariably include parameters that are not well informed by existing data. Therefore, a key component of model development is the choice of parameter values. Microsimulation model parameter values are selected to reproduce expected or known results though the process of model calibration. Calibration may be done by perturbing model parameters one at a time or by using a search algorithm. As an alternative, we propose a Bayesian method to calibrate microsimulation models that uses Markov chain Monte Carlo. We show that this approach converges to the target distribution and use a simulation study to demonstrate its finite-sample performance. Although computationally intensive, this approach has several advantages over previously proposed methods, including the use of statistical criteria to select parameter values, simultaneous calibration of multiple parameters to multiple data sources, incorporation of information via prior distributions, description of parameter identifiability, and the ability to obtain interval estimates of model parameters. We develop a microsimulation model for colorectal cancer and use our proposed method to calibrate model parameters. The microsimulation model provides a good fit to the calibration data. We find evidence that some parameters are identified primarily through prior distributions. Our results underscore the need to incorporate multiple sources of variability (i.e., due to calibration data, unknown parameters, and estimated parameters and predicted values) when calibrating and applying microsimulation models.
In order to assess the time at which the distant metastases were initiated, a model has been developed to simulate the natural history of human breast cancer. The metastasis appearance curves were fitted to those observed for tumours of various sizes among the 2648 patients treated at the Institut Gustave Roussy from 1954 to 1972. The model assumes that metastases are initiated when the tumour reaches a threshold volume (distribution of this volume was estimated in a previous article). Two patterns of growth were considered: exponential and Gompertzian. Distributions of tumour and metastases doubling times are fixed according to the literature. A relationship between tumour and metastasis doubling time is estimated. Simulations were used to optimize metastases growth duration as a function of the metastasis doubling time. The ages of the metastases at tumour diagnosis are calculated. With exponential growth, it was necessary to introduce correlations to obtain a satisfactory fit of the metastases appearance curves: between the tumour volume at diagnosis and the doubling time (R1 = -0.3), and between the tumour volume at metastasis initiation and the doubling time (R2 = 0.3). The growth duration of the metastases before their detection was found to equal about 18 metastases doubling times at detection and the mean ratio between the doubling time of a tumour and its metastases equal to 2.2. With Gompertzian growth, it was impossible to adjust satisfactorily the proportions of metastases at diagnosis as a function of the primary tumour volume. However, when we ignore this, the best fit was obtained with the duration of metastases growth before detection was about the same as for exponential growth. With either growth pattern, the model predicts that the proportion of patients with metastases would be reduced by ~3% if the primary tumours were treated 12 months earlier. This prediction is consistent with the results of the screening programs for breast cancer.
The relationship between the size of the primary tumour upon initial treatment and the incidence of distant metastasis during the course of the disease was investigated using data from 2648 breast cancers treated at the Institut Gustave Roussy between 1954 and 1972. This analysis suggests the existence for each tumour of a critical volume (threshold) at which the first remote metastasis is initiated. The correlation between the size of the primary tumour and the probability of metastatic dissemination was assessed as well as the influence on this correlation of two prognostic indicators: histological grade and number of involved lymph nodes. It was found that the threshold volume is strongly correlated with the number of involved lymph nodes and the histological grading.
Markov models are useful when a decision problem involves risk that is continuous over time, when the timing of events is important, and when important events may happen more than once. Representing such clinical settings with conventional decision trees is difficult and may require unrealistic simplifying assumptions. Markov models assume that a patient is always in one of a finite number of discrete health states, called Markov states. All events are represented as transitions from one state to another. A Markov model may be evaluated by matrix algebra, as a cohort simulation, or as a Monte Carlo simulation. A newer representation of Markov models, the Markov-cycle tree, uses a tree representation of clinical events and may be evaluated either as a cohort simulation or as a Monte Carlo simulation. The ability of the Markov model to represent repetitive events and the time dependence of both probabilities and utilities allows for more accurate representation of clinical settings that involve these issues.
Hazelton, W. D., Luebeck, E. G., Heidenreich, W. F. and Moolgavkar, S. H. Analysis of a Historical Cohort of Chinese Tin Miners with Arsenic, Radon, Cigarette Smoke, and Pipe Smoke Exposures Using the Biologically Based Two-Stage Clonal Expansion Model. Radiat. Res. 156, 78-94 (2001).The two-stage clonal expansion model is used to analyze lung cancer mortality in a cohort of Yunnan tin miners based on individual histories with multiple exposures to arsenic, radon, cigarette smoke, and pipe smoke. Advances in methodology include the use of nested dose-response models for the parameters of the two-stage clonal expansion model, calculation of attributable risks for all exposure combinations, use of both a fixed lag and a gamma distribution to represent the time between generation of the first malignant cell and death from lung cancer, and scaling of biological parameters allowed by parameter identifiability. The cohort consists of 12,011 males working for the Yunnan Tin Corporation, with complete exposure records, who were initially surveyed in 1976 and followed through 1988. Tobacco and arsenic dominate the attributable risk for lung cancer. Of 842 lung cancer deaths, 21.4% are attributable to tobacco alone, 19.7% to a combination of tobacco and arsenic, 15.8% to arsenic alone, 11% to a combination of arsenic and radon, 9.2% to a combination of tobacco and radon, 8.7% to combination of arsenic, tobacco and radon, 5.5% to radon alone, and 8.7% to background. The models indicate that arsenic, radon and tobacco increase cell division, death and malignant conversion of initiated cells, but with significant differences in net cell proliferation rates in response to the different exposures. Smoking a bamboo water pipe or a Chinese long-stem pipe appears to confer less risk than cigarette use, given equivalent tobacco consumption.
Experimental evidence indicates that tobacco smoke acts both as an initiator and a promoter in lung carcinogenesis. We used the two-stage clonal expansion model incorporating the ideas of initiation, promotion, and malignant conversion to analyze lung cancer mortality in three large cohorts, the British Doctors' cohort and the two American Cancer Society cohorts, to determine how smoking habits influence age-specific lung cancer rates via these mechanisms. Likelihood ratio tests indicate that smoking-related promotion is the dominant model mechanism associated with lung cancer mortality in all cohorts. Smoking-related initiation is less important than promotion but interacts synergistically with it. Although no information on ex-smokers is available in these data, the model with estimated variables can be used to project risks among ex-smokers. These projected risks are in good agreement with the risk among ex-smokers derived from other studies. We present 10-year projected risks for current and former smokers adjusted for competing causes of mortality. The importance of smoking duration on lung cancer risk in these cohorts is a direct consequence of promotion. Intervention and treatment strategies should focus on promotion as the primary etiologic mechanism in lung carcinogenesis.
Although previous empirical studies have shown that tobacco control policies are effective at reducing smoking rates, such studies have proven of limited effectiveness in distinguishing how the effect of policies depend on the other policies in place, the length of adjustment period, the way the policy is implemented, and the demographic groups considered. An alternative and complementary approach to purely statistical equations is simulation models. We describe the SimSmoke simulation model and how we used it to assess tobacco control policy in a specific case study. Simulation models are not only useful for policy prediction and planning but also may help to broaden our understanding of the role of different public health policies within a complex, dynamic social system.
Objectives. The authors develop a simulation model to predict the effects on quit rates and cost-effectiveness of different smoking treatment policies.Methods. A decision theoretic model of quit behavior is first developed that incorporates the decision to quit and the choice of treatment. A policy model then examines the effect on quit attempts and quit rates of policies to cover the costs of different combinations of treatments and to require health care providers to conduct brief interventions. The model incorporates substitution between treatments and effects of policies on treatment effectiveness. The cost per quit is also calculated for each policy.Results. The model of quit behavior predicts a 1-year quit rate of 4.5% for the population of smokers. The policy model predicts a 37% increase in quit rates from a policy that combines mandated brief interventions with coverage of all proven tobacco treatments. Smaller effects are predicted from policies that provide more restricted coverage of treatments, especially those limited to behavioral treatment. Payments for brief interventions alone increase quit rates by about 7%. Brief intervention and behavioral therapy policies had lower costs per quit but yield substantially fewer additional quits than policies that cover pharmacotherapy. There is, however, considerable variation around these estimates depending on assumptions about the effects of policy on treatment use, substitution between treatments, and treatment effectiveness.Conclusion. Tobacco treatment policies, especially those with broad and flexible coverage, have the potential to substantially increase smoking quit rates. However, further research is needed on the effect of payment policies on the use and effectiveness of tobacco treatments.
With the shift toward individualized treatment, cost-effectiveness models need to incorporate patient and tumor characteristics that may be relevant to treatment planning. In this study, we used multistate statistical modeling to inform a microsimulation model for cost-effectiveness analysis of individualized radiotherapy in lung cancer. The model tracks clinical events over time and takes patient and tumor features into account. Four clinical states were included in the model: alive without progression, local recurrence, metastasis, and death. Individual patients were simulated by repeatedly sampling a patient profile, consisting of patient and tumor characteristics. The transitioning of patients between the health states is governed by personalized time-dependent hazard rates, which were obtained from multistate statistical modeling (MSSM). The model simulations for both the individualized and conventional radiotherapy strategies demonstrated internal and external validity. Therefore, MSSM is a useful technique for obtaining the correlated individualized transition rates that are required for the quantification of a microsimulation model. Moreover, we have used the hazard ratios, their 95% confidence intervals, and their covariance to quantify the parameter uncertainty of the model in a correlated way. The obtained model will be used to evaluate the cost-effectiveness of individualized radiotherapy treatment planning, including the uncertainty of input parameters. We discuss the model-building process and the strengths and weaknesses of using MSSM in a microsimulation model for individualized radiotherapy in lung cancer.
© The Author(s) 2015.
The National Lung Screening Trial (NLST) demonstrated that low-dose computed tomography screening is an effective way of reducing lung cancer (LC) mortality. However, optimal screening strategies have not been determined to date and it is uncertain whether lighter smokers than those examined in the NLST may also benefit from screening. To address these questions, it is necessary to first develop LC natural history models that can reproduce NLST outcomes and simulate screening programs at the population level.
Five independent LC screening models were developed using common inputs and calibration targets derived from the NLST and the Prostate, Lung, Colorectal and Ovarian Cancer Screening Trial (PLCO). Imputation of missing information regarding smoking, histology, and stage of disease for a small percentage of individuals and diagnosed LCs in both trials was performed. Models were calibrated to LC incidence, mortality, or both outcomes simultaneously.
Initially, all models were calibrated to the NLST and validated against PLCO. Models were found to validate well against individuals in PLCO who would have been eligible for the NLST. However, all models required further calibration to PLCO to adequately capture LC outcomes in PLCO never-smokers and light smokers. Final versions of all models produced incidence and mortality outcomes in the presence and absence of screening that were consistent with both trials.
The authors developed 5 distinct LC screening simulation models based on the evidence in the NLST and PLCO. The results of their analyses demonstrated that the NLST and PLCO have produced consistent results. The resulting models can be important tools to generate additional evidence to determine the effectiveness of lung cancer screening strategies using low-dose computed tomography. Cancer 2014. © 2014 American Cancer Society.
As a member of the Cancer Intervention and Surveillance Modeling Network (CISNET), the lung cancer (LC) group at Fred Hutchinson Cancer Research Center (FHCRC) developed a model for evaluating U.S. lung cancer mortality trends and the impact of changing tobacco consumption. Model components include a biologically based two-stage clonal expansion (TSCE) model; a smoking simulator to generate smoking histories and other cause mortality; and adjustments for period and birth cohort to improve calibration to U.S. LC mortality. The TSCE model was first calibrated to five substantial cohorts: British doctors, American Cancer Society CPS-I and CPS-II, Health Professionals' Follow-Up Study (HPFS), and Nurses' Health Study (NHS). The NHS and HPFS cohorts included the most detailed smoking histories and were chosen to represent the effects of smoking on U.S. LC mortality. The calibrated TSCE model and smoking simulator were used to simulate U.S. LC mortality. Further adjustments were necessary to account for unknown factors. This provided excellent fits between simulated and observed U.S. LC mortality for ages 30-84 and calendar years 1975-2000. The FHCRC LC model may be used to study the effects of public health information on U.S. LC trends and the impact of tobacco control policy. For example, we estimated that over 500,000 males and 200,000 females avoided LC death between 1975 and 2000 due to increasing awareness since the mid 1950s of the harmful effects of smoking. We estimated that 1.1 million male and 0.6 million female LC deaths were avoidable if smokers quit smoking in 1965.
The Rice-MD Anderson group uses a two-stage clonal expansion (TSCE) model of lung cancer mortality calibrated to a combination of MD Anderson case-control data on smoking histories and lung cancer mortality/incidence rate data collected from prospective cohorts in order to predict risk of lung cancer. This model is used to simulate lung cancer mortality in the U.S. population under the three scenarios of CISNET lung group's smoking base case project in order to estimate the effect of tobacco control policy on lung cancer mortality rates. Simulation results show that tobacco control policies have achieved 35% of the reduction in lung cancer mortality that would have resulted from cessation of all smoking in 1965.
The natural history model underlying the MGH Lung Cancer Policy Model (LCPM) does not include the two-stage clonal expansion model employed in other CISNET lung models. We used the LCPM to predict numbers of U.S. lung cancer deaths for ages 30-84 between 1975 and 2000 under four scenarios as part of the comparative modeling analysis described in this issue. The LCPM is a comprehensive microsimulation model of lung cancer development, progression, detection, treatment, and survival. Individual-level patient histories are aggregated to estimate cohort or population-level outcomes. Lung cancer states are defined according to underlying disease variables, test results, and clinical events. By simulating detailed clinical procedures, the LCPM can predict benefits and harms attributable to a variety of patient management practices, including annual screening programs. Under the scenario of observed smoking patterns, predicted numbers of deaths from the calibrated LCPM were within 2% of observed over all years (1975-2000). The LCPM estimated that historical tobacco control policies achieved 28.6% (25.2% in men, 30.5% in women) of the potential reduction in U.S. lung cancer deaths had smoking had been eliminated entirely. The hypothetical adoption in 1975 of annual helical CT screening of all persons aged 55-74 with at least 30 pack-years of cigarette exposure to historical tobacco control would have yielded a proportion realized of 39.0% (42.0% in men, 33.3% in women). The adoption of annual screening would have prevented less than half as many lung cancer deaths as the elimination of cigarette smoking.
We discuss the hazard function of the two-mutation clonal expansion model with time-dependent parameters, with particular emphasis on identifiability of the parameters. We explicitly construct identifiable parameter combinations, and illustrate the properties of the hazard function under perturbations of the underlying biological parameters.
By permission of MIT Press reproduced below is the full text of Orcutt G (1957) ‘A new type of socio-economic system’, Review of Economics and Statistics, 39(2), 116-123.
The Colorectal Cancer Simulated Population model for Incidence and Natural history (CRC-SPIN) is a new microsimulation model for the natural history of colorectal cancer that can be used for comparative effectiveness studies of colorectal cancer screening modalities.
CRC-SPIN simulates individual event histories associated with colorectal cancer, based on the adenoma-carcinoma sequence: adenoma initiation and growth, development of preclinical invasive colorectal cancer, development of clinically detectable colorectal cancer, death from colorectal cancer, and death from other causes. We present the CRC-SPIN structure and parameters, data used for model calibration, and model validation. We also provide basic model outputs to further describe CRC-SPIN, including annual transition probabilities between various disease states and dwell times. We conclude with a simple application that predicts the impact of a one-time colonoscopy at age 50 on the incidence of colorectal cancer assuming three different operating characteristics for colonoscopy.
CRC-SPIN provides good prediction of both the calibration and the validation data. Using CRC-SPIN, we predict that a one-time colonoscopy greatly reduces colorectal cancer incidence over the subsequent 35 years.
CRC-SPIN is a valuable new tool for combining expert opinion with observational and experimental results to predict the comparative effectiveness of alternative colorectal cancer screening modalities.
Microsimulation models such as CRC-SPIN can serve as a bridge between screening and treatment studies and health policy decisions by predicting the comparative effectiveness of different interventions. As such, it is critical to publish model descriptions that provide insight into underlying assumptions along with validation studies showing model performance.
"September 2005." Thesis (Ph.D., Committee on Higher Degrees in Health Policy)--Harvard University, 2005. Includes bibliographical references.
A two-mutation model for carcinogenesis is reviewed. General principles in fitting the model to epidemiologic and experimental data are discussed, and some examples are given. A general solution to the model with time-dependent parameters is developed, and its use is illustrated by application to data from an experiment in which rats exposed to radon developed lung tumors.
Population based data on smoking history derived from NCHS surveys were used to develop a model for lung cancer incidence in Connecticut. Trends in smoking prevalence suggest that, while the prevalence in men increased earlier than women, more male smokers have quit than their female counterparts. These trends in smoking prevalence suggest striking gender differences in a period effect for the smoking prevalence. Estimates of the proportion of current smokers, ex-smokers, and the mean duration of smoking were used in a model for the lung cancer incidence rates. The form for the relationship between smoking history and the incidence rate for these subgroups was based on information from cohort studies. The models represented a mixture of the smoking subgroups where the effect of smoking was considered to be either a multiplicative effect on the underlying age distribution, or a separate effect in which the level of exposure was the sole contribution to risk among smokers. The multiplicative model explained more than 80 per cent of the deviance for the period and cohort effects, while the non-multiplicative model could only account for trends in females. Hence, these results suggest that a sizeable portion of the period and cohort contributions to the lung cancer incidence trends in Connecticut can be attributed to the multiplicative model that utilizes this smoking information, although the lack of more detailed information is a limiting factor in developing the model.
To understand cancer aetiology better, epidemiologists often try to investigate the time trends in disease incidence with year of diagnosis (period) and birth cohort. Unfortunately, one cannot identify these factors uniquely in the usual regression model owing to a linear dependence between age, period and cohort, so that one requires additional information about the underlying biology of the disease. Carcinogenesis models provide one type of information that can result in a unique set of parameters for the effects of age, period and cohort. We use the multistage carcinogenesis model and its extensions to obtain a unique set of parameters for an age-period-cohort model of lung cancer trends of Connecticut males and females from 1935 to 1988. Some of these models do not seem to provide a reasonable set of model parameters, but we found that a model that included second-order terms and a multistage mixture model both gave a good fit to the data and realistic parameter estimates.
Tobacco was introduced into Europe from America at the end of the fifteen century. At first used primarily for medicinal purposes it came to be burnt in pipes for pleasure on a large scale nearly 100 years later, at first in England and subsequently in Europe and throughout the world. Pipe smoking gave way to the use of tobacco as snuff and, in turn, to cigars and cigarettes at different times in different countries until cigarette smoking became the dominant form in most of the developed world between the two world wars. Societies were formed to discourage smoking at the beginning of the century in several countries, but they had little success except in Germany where they were officially supported by the government after the Nazis seized power. In retrospect it can now be seen that medical evidence of the harm done by smoking has been accumulating for 200 years, at first in relation to cancers of the lip and mouth, and then in relation to vascular disease and cancer of the lung. The evidence was generally ignored until five case-control studies relating smoking to the development of lung cancer were published in 1950. These stimulated much research, including the conduct of cohort studies, which, by the late 1950s, were beginning to show that smoking was associated with the development of many other diseases as well. The interpretation that smoking caused these various diseases was vigorously debated for some years but came to be generally accepted in respect of lung cancer by the late 1950s and of many other diseases in the subsequent two decades. Cigarette smoking has now been found to be positively associated with nearly 40 diseases or causes of death and to be negatively associated with eight or nine more. In some instances the positive associations are largely or wholly due to confounding, but the great majority have been shown to be causal in character. The few diseases negatively associated with smoking are for the most part rare or nonfatal and their impact on disease incidence and mortality as a result of smoking is less than 1% of the excess of other diseases that are caused by smoking. The most recent observations show that continued cigarette smoking throughout adult life doubles age-specific mortality rates, nearly trebling them in late middle age. All the diseases related to smoking that cause large numbers of deaths should now have been discovered, but further nonfatal diseases may remain to be revealed by cohort studies that are able to link individuals' morbidity data with their personal characteristics.
The sequencing of treatment for early breast cancer is controversial. The purpose of this study was to quantify the risk of delaying surgery, using estimates of the frequency of first metastases from breast primary tumors.
The probability that 560 (node-negative), 657 (with one to three positive nodes), and 505 (with more than three positive nodes) women treated without adjuvant chemotherapy would be free of distant disease at presentation was fit to a mathematical model of the seeding of distant metastases and combined with estimates of the growth rate to calculate the frequency of first distant disseminations per month.
Frequencies of first distant metastases were approximately 1% to 2% per month, 2% to 4% per month, and 3% to 6% per month in T1 patients who were node-negative, had one to three positive nodes, or more than three positive nodes, respectively. As a result, the typical patient with T1 disease, who has a 70% to 80% chance of being free of distant disease, runs a 1% to 4% risk of distant dissemination for each month surgery is delayed. Assuming a 30% reduction in mortality caused by adjuvant chemotherapy, the model predicts that T1 patients treated with neoadjuvant chemotherapy would potentially have a higher rate of distant metastasis development than those treated with an initial surgical resection followed by adjuvant chemotherapy.
We formulate the hypothesis that optimal sequencing of surgery and systemic treatment of breast cancer may be size-dependent, with a disadvantage or no benefit from neoadjuvant treatment for T1 patients but an increasing benefit with increasing size of the primary tumor.
Invasive breast cancer is commonly staged as local, regional or distant disease. We present a stochastic model of the natural history of invasive breast cancer that quantifies (1) the relative rate that the disease transitions from the local, regional to distant stages, (2) the tumour volume at the stage transitions and (3) the impact of symptom-prompted detection on the tumour size and stage of invasive breast cancer in a population not screened by mammography. By symptom-prompted detection, we refer to tumour detection that results when symptoms appear that prompt the patient to seek clinical care. The model assumes exponential tumour growth and volume-dependent hazard functions for the times to symptomatic detection and stage transitions. Maximum likelihood parameter estimates are obtained based on SEER data on the tumour size and stage of invasive breast cancer from patients who were symptomatically detected in the absence of screening mammography. Our results indicate that the rate of symptom-prompted detection is similar to the rate of transition from the local to regional stage and an order of magnitude larger than the rate of transition from the regional to distant stage. We demonstrate that, in the even absence of screening mammography, symptom-prompted detection has a large effect on reducing the occurrence of distant staged disease at initial diagnosis.
As newer therapies for lung cancer are being explored it becomes more important to understand the natural history of lung cancer. A systematic review of the data shows that untreated lung cancer is almost uniformly rapidly fatal, even if it is stage I. Analysis of data regarding tumor volume doubling times shows that conventionally detected lung cancers have short mean doubling times, and only a small proportion with very long doubling times. Lung cancers found during the course of a CT screening program have markedly longer mean doubling times and a substantially greater proportion with very long doubling times (>400 days). Models of tumor growth, however, are not understood well enough to use the observed doubling time to predict length of survival without treatment.
Multi-state statistical modelling to quantify an individual-based micro simulation model for radiotherapy treatment in lung cancer patients
- M L Bongers
- D De Ruysscher
- C Oberije
- P Lambin
- C A Uyl-De Groot
- V M H Coupe
Bongers, M. L., De Ruysscher, D., Oberije, C., Lambin, P., Uyl-de Groot, C. A., & Coupe, V. M. H.
(2013). Multi-state statistical modelling to quantify an individual-based micro simulation model
for radiotherapy treatment in lung cancer patients. Value in Health, 16 (7), A325-A325.
Statistical methods in micro-simulation modeling: Calibration and predictive accuracy (Doctoral dissertation
- S A Chrysanthopoulou
Chrysanthopoulou, S. A. (2013). Statistical methods in micro-simulation modeling: Calibration
and predictive accuracy (Doctoral dissertation, Brown University). Retrieved from https://
repository.library.brown.edu/studio/item/bdr:386153/ (Biostatistics Dissertations, Brown Digital Repository)
MILC: Microsimulation lung cancer (MILC) model [Computer software manual
- S A Chrysanthopoulou
Chrysanthopoulou, S. A. (2014). MILC: Microsimulation lung cancer (MILC) model [Computer
software manual]. Retrieved from http://CRAN.R-project.org/package=MILC (R
package version 1.0)
Comparative Analysis of Calibration Methods for Microsimulation Models
- S A Chrysanthopoulou
Chrysanthopoulou, S. A. (2017). Comparative Analysis of Calibration Methods for Microsimulation Models.
(Manuscript in preparation)
Assessing the Predictive Accuracy of Microsimulation Models
- S A Chrysanthopoulou
Chrysanthopoulou, S. A. (2018). Assessing the Predictive Accuracy of Microsimulation Models.
(Manuscript in preparation)
Parameter estimation in mathematical models of lung cancer (Unpublished doctoral dissertation)
- D L Goldwasser
Goldwasser, D. L. (2009). Parameter estimation in mathematical models of lung cancer (Unpublished
doctoral dissertation). RICE University.
Toward a new comprehensive international health and health care policy decision support tool
- J Oderkirk
- F Sassi
- M Cecchini
Oderkirk, J., Sassi, F., & Cecchini, M. (2012). Toward a new comprehensive international health and
health care policy decision support tool. OECD.
Dynamic microsimulation models for health outcomes
- C M Rutter
- A M Zaslavsky
- E J Feuer
Rutter, C. M., Zaslavsky, A. M., & Feuer, E. J. (2011). Dynamic microsimulation models for health
outcomes. Medical Decision Making, 31(1), 10-18.
Growth kinetics of tumours : cell population kinetics in relation to the growth and treatment of cancer
- G G Steel
Steel, G. G. (1977). Growth kinetics of tumours : cell population kinetics in relation to the growth and
treatment of cancer. Oxford: Clarendon Press.
International agency for research on cancer. Tobacco smoking
IARC. (1986). International agency for research on cancer. Tobacco smoking. IARC Monographs on
the Evaluation of Carcinogenic Risks to Humans, Vol. 38.