《MarekKimmel》由会员分享，可在线阅读，更多相关《MarekKimmel（23页珍藏版）》请在金锄头文库上搜索。
1、Modeling Lung Cancer,Olga Y. GorlovaMD Anderson Cancer CenterMarek KimmelRice University,Natural history, detection and intervention Optimization of screening strategies, prediction of effectiveness, costsHealth informatics Comprehensive follow-up and “intelligent registry of cases”, understanding p
2、ooling of dataNo viable alternatives Ultimately, RCT unethical, too slow to respond to technology changes and vulnerable to “fatigue” problems,Challenges for modeling,Visualization Detection of pre-cancerous lesions, false positives Multiple gene expression Molecular classification of tumors with re
3、spect to progressionGenetic screening Variable risk of LC depending on genetic makeup, stratification of high-risk population,Mayo trial: Design, flaws and outcome,Design 10,000 high-risk males randomized into screening and control. Screenees: examined each 4 months by Xray and sputum cytology (for
4、6 years) and instantly treated.Flaws High contamination of controls by “voluntary screening” (controls took chest Xray each 2 years). Noncompliance of screenees (60% compliance rate).Outcome Excess of early cases in the screened group (attributed to biases). No significant difference in mortality.,S
5、tatistical model of lung cancer progression, detection and treatment,Approach To build a mathematical/computer model, fit it to data, and generate predictions.Key parameters = mean duration of early LCp = probability of detection of early LC by screeningc = curability of early LC (Flehinger and Kimm
6、el, Biometrics, 1987, 1990),Lung cancer model,Sequence of events in the model of cancer and screening. In the case depicted, screenings begin before cancer is present. The cancer is missed by two screens, but detected while still in early stage.,No screening,Periodic screening,Modeling Mayo Lung Pro
7、ject,Assume most favorable case fitting data.Long-term population reduction in mortality around 10%.Model 6 years of screening (500 simulation runs)Distributions of numbers of lung cancer deaths in study vs. control indistinguishable (practically 0 power)(based on Flehinger et al. 1992),Marcus et al
8、 (2000) Follow-up Study,Data on mortality and survival for follow-up period extended to median 20.5 years.No mortality difference found between study and control.Some difference in survival between study and control.Conclusion: No benefit from screening, difference in survival attributable to “clini
9、cally irrelevant lesions found in screening”.,Modeling MLP with extended follow-up,Again, assume most favorable case fitting data.Model 6 years of screening + 20 years of follow-up.Number of deaths in the model match the numbers in Marcus et al. (2000) Distributions of numbers of lung cancer deaths
10、in study vs. control indistinguishable,Conclusion: Extended follow-up does not help in increasing power.,MLP: Extended screening vs. extended follow-up,Survival in extended follow-up,(a) Modeled, based on no benefit from screening (difference totally attributable to biases)(b) Modeled, based on maxi
11、mum benefit from screening (c) Marcus et al. (2000),Reduction of mortality with increaseddetectability and/or curability,p=0.245c=0.35,p=0.8c=0.35,p=0.245c=0.8,p=0.8c=0.8,Years of screening,Number of LC deaths,Conclusions,Extended follow-up after a screening study not likely to add powerExtended scr
12、eening is required to gather benefits of early detection and cure (differential mortality reaches a plateau after some time)Assertion that only “clinically irrelevant” cases are picked up by screening not consistent with survival dataIncreased detection and curability make screening worthwhile.,Mode
13、ling earliest stages of cancer,Role of tumor size in modeling,Estimated distribution function of tumor size at the point of distant metastasis (s) (Kimmel and Flehinger, Biometrics 1990),Supplement follows,Early detection of chronic diseases by periodic screening,General ideaIdentify the “at risk” p
14、opulation (smokers in LC case)Apply an “early detection” procedure (chest Xray in LC case), periodically, among the members of the “at risk” populationBy treating the early cases discovered this way, a much higher cure rate is assured than that of symptomatic cases.Is all this real?,Screening biases
15、,Lead-time bias: Early detection extends apparent survival even if early detection does not improve treatment (hence, survival curves misleading!),Screening biases,Length-biased sampling: Screening tends to detect “slower” and hence more indolent cases (effect like in observing the renewal process).
16、 If treated, these cases inflate cure estimates.,Randomized clinical trial (RCT),A possible (?) way to avoid biases,Lung cancer screening trials, Mayo, Hopkins, Sloan-Kettering,1970-1983,Clinical trials in leading medical institutions, 10,000 high-risk individuals each.Purpose: To determine if annual screening by chest Xray and sputum cytology reduces mortality from lung cancer.In Mayo trial, power of 80% to detect 50% reduction in mortality at 5% significance level.,
