Overdiagnosis in Mammography Screening Jean Ching-Yuan Fann, Huei-Shian Tsau, Chen-Yang Hsu, King-Jen Chang, Amy Ming-Fang Yen, Cheng-Ping Yu, Sam Li-Sheng Chen, Wen-Hung Kuo, László Tabár, Sherry Chiu, Hsiu-Hsi Chen -Sep-28 Outline Mammographic Screening for Breast Cancer Fallacy on Overdiagnosis Overdiagnosis in Taiwanese Randomized Controlled Trial Methodology for Estimating Overdiagnosis Personalized Probabilistic Cost-Effectiveness Analysis 2
Meta-analyses: UK Independent 22 Lancet Average effect: 2% mortality reduction 3 4 2
Cumulative Incidence (per ) Fallacy in BC mass screening. Short follow-up time: without lead-time consideration 2. Breast Cancer mixed: diagnosed before screening program, but died after program implementation With an average follow-up of 2.2 years BMJ, 2 Mixed up lead-time and over-detection Jørgensen et al., 29 Lead-time period (lead-time bias) 5 Overdiagnosis with mammography in Taiwan based on the Taiwanese randomized controlled trial for young women Eligible Population Randomization N=2,4 N=2,87 N=39,563 M U M U U M U M Control Arm M: Mammography U: Ultrasound 9 8 7 6 5 4 3 2 Control Group Screen Group (M-U + U-M) RR=.3 (.94-.35) Time since randomization (month) 6 3
Overdiagnosis with mammography in Taiwan based on the Taiwanese Population-based service screening 26 JAMA Oncology Total Incidence of breast cancer Mammography vs CBE: RR=.3 (95% CI:.8-.8) Over-detection: 3% 7 4
Methodology for Estimating Overdiagnosis. Graphic method 2. Zero-inflated model 3. Coxian Phase-Type Markov Process 9. Graphic method Curved method by comparing cumulative incidence of cancer Non-advanced cancer cancer Upper limit All over-detection arise from Non-advanced cancer (Long follow up time) advanced cancer Chen et al.,27 Screen Works! Lower limit All detected cancer became advanced cancer(no over-detection) 5
Survival probabiltiy Assessing overdetection in breast cancer screening using data on randomized controlled trial Chen et al.,27 Medicine Follow-up time 2. Zero-inflated model Survival of Breast Cancer, Darlana, Sweden Without consideration of over-diagnosis.9 9%.8.7.6.5 Survival adjusted for prognostic factors.4.3.2. 2 3 4 5 6 7 8 9 2 3 4 5 2 Year since diagnosis 6
Survival probability Zero-inflated Poisson regression model and overdiagnosis rate Variable Count part RR/OR (95% CI) RR P-value Intercept Size, mm.5-4 vs. -9 3.69(.76-8.) 5-9 vs. -9 3.85(.8-8.53) 2-29 vs. -9.26(2.27-46.33) 3+ vs. -9 9.45(2.-44.49) Node (+) vs. (-) 2.4(.3-4.45).5 Grade 3 vs /2.62(.94-2.79).8 Surgery MA vs. BCS.92(.95-3.88).7 Triple Negative Yes vs No 2.49(.36-4.59).3 Chemotherapy Yes vs. No.79(.42-.47).456 Radiotherapy Yes vs. No.23(.6-2.53).568 Tamoxifen Yes vs. No.95(.94-.64).847 Zero part OR Intercept Detection mode.4 SD vs. RF 2.38(.97-5.85) IC vs. RF.23(.48-3.7) = 56.4% SD: 66.4% Overdiagnosis, 8.9% IC: 5.5% Awareness, 2.9% RE: 45.4% Treatment effect Pr(Y = y; μ, π) π + π e μ when y = =.9.8.7.6.5.4.3.2. π e μ μ y y! when y =,2,3 Count Part: Poisson regression for μ Zero Part: Logistic regression for π Survival adjusted for prognostic factors Survival considering Overdiagnosis 9% 75% 2 3 4 5 6 7 8 9 2 3 4 5 3 Year since diagnosis 3. Coxian Phase-Type Markov Process Applying the concept of cured model: S t = S P t π + S NP t π For exponentially distributed random variable exp α t = π exp α P t + π π = exp α t exp α P t exp α P t 4 7
Estimated natural history of breast cancer with and without consideration of over-detection, Swedish Two-County Trial (Kopparberg) 977-985 Transition rate π = 2.6% MST: mean sojourn time 5 Probabilistic CEA of Personalized Breast Cancer Screening Population risk stratification for trade-off between harm and benefit Time preference for screening policy and outcome 6 8
Risk stratification: The recommend age to begin screening and inter-screening interval for screening by percentiles of risk score Age to begin screening 75 Inter-screening Interval Intensive 65 55 45 35 25 - -2 2-3 3-4 4-5 5-6 6-7 7-8 8-9 9- (Percentile) 2 4 6 8 2 Early commencing Low Risk High Risk Economic Evaluation Acceptability curve of primary and secondary breast cancer prevention for non-brca Carrier.9 2 GDP.8.7.6.5.4.3.2 Annual mammography Biennial mammography Triennial mammography Risk-based screening interval. 3 3 32 33 34 35 36 37 38 39 4 4 42 43 44 45 46 47 48 49 5 5 8 9
Acceptability curve of primary and secondary prevention of breast cancer for BRCA-carrier women.9.8.7 GDP 2 GDP 3 GDP Annual Mammography Annual MRI+ mammography.6.5.4.3.2. Preventive Surgery, Annual mammography Preventive Surgery, Annual MRI+ mammography Chemoprevention, annual mammography Chemoprevention, annual MRI +mammography 5 9 3 7 2 25 29 33 37 4 45 49 53 57 6 65 69 9 Conclusion The estimated proportion of over-diagnosis cases is affected by lead-time, sensitivity, and follow-up time, which causes the large disparity of over-detection across studies. - Methodological flaws Use high-quality design-based study and model-based approach Probabilistic CEA for personalized screening policy is strongly recommended 2
Thanks for your attention!