Lead Time Bias as a Factor in Mammography Detected Invasive Breast Cancer Survival in an Institutional Cohort

Background: Lead time, the interval between screen detection and when a disease would have become clinically evident, is commonly cited to explain longer survival times in mammography detected breast cancer cases (BC). Methods: An institutional retrospective cohort study of BC outcomes related to detection method (mammography (MamD) vs. patient (PtD)). Cases were rst primary invasive stage I-III BC, age 40-74 years (n = 6603), 1999-2016. Survival time was divided into 1) distant disease-free interval (DDFI) and 2) distant disease-specic survival (DDSS) as two separate time interval outcomes. We measured statistical association between detection method and diagnostic, treatment and outcome variables using bivariate comparisons, Cox proportional hazards analyses and mean comparisons. Outcomes were distant recurrence (n=422), DDFI and DDSS. Results: 39% of cases were PtD (n = 2566) and 61% were MamD (n = 4037). MamD cases had a higher percentage of Stage I tumors [MamD 69% stage I vs. PtD 31%, p<.001]. Rate of distant recurrence was 11% among PtD BC cases (n=289) vs. 3% of MamD (n=133) (p<.001). Order of factor entry into the distant recurrence time interval (DDFI) model was 1) TNM stage (p<.001), 2) HR/HER2 status (p<.001), 3) histologic grade (p=.005) and 4) detection method (p<.001). Unadjusted PtD DDFI mean time was 4.34 years and MamD 5.52 years (p<.001) however when stratied by stage, the most signicant factor relative to distant recurrence, there was no signicant difference between PtD and MamD BC. Distant disease specic survival time did not differ by detection method. Conclusion: We observed breast cancer survival differential lead time to be a function of stage at diagnosis and tumor characteristics with marginal contribution of detection method. Patient and mammography detected breast cancer time to distant recurrence did not differ stratied by stage indicating survival difference is more likely related to early diagnosis than lead time bias. Lead time bias associated with breast cancer detection method appears to have marginal inuence on survival in the current diagnostic and treatment era.


Introduction
The incidence of recurrent metastatic breast cancer (rMBC) has decreased in recent years coincident with an improvement in breast cancer survival in part due to successful adjuvant therapy improvement for stage I-III disease. 1,2,3 Debate and analysis continue about the relative contribution of early detection of breast cancer by mammography screening to improved survival. From reports of national mammography screening program surveillance reports, mammography detected tumors are more often smaller and lower stage with better survival but these reports lack comparative evidence as to whether this is a real contributor to improved survival or the result of bias. 4 Lead time bias is the time between screening detection and when the disease would have become clinically evident without screening. Lead time interval when added to the time over which evident disease progresses, makes it appear that screen detected cases have longer survival than if screening had not taken place. 5,6,7,8,9 We are now in a time of accepted validity for mammography screening with evidence-based guidelines adopted and promoted in the United States and Europe. 10,11,12 Mammography screening is not institutionalized in the United States where it is an opportunistic choice based on health care access, screening guideline knowledge, insurance coverage and care giver recommendation. 13 Timing and incidence of invasive breast cancer distant disease recurrence provides an opportunity to measure lead time by comparing time to distant recurrence after initial diagnosis and post recurrence survival as a function of detection method. In this study of invasive breast cancer in a retrospective institutional cohort, we reviewed breast cancer characteristics by method of detection with time to and incidence of distant disease recurrence and death as the outcomes of interest to assess contribution of lead time to survival.

Methods
To assess the contribution of lead time to survival among mammography detected BC cases we conducted lead time analysis comparing mammography (MamD) to patient detected (PtD) BC using time to distant recurrence as the rst interval (DDFI) and time from distant recurrence to last follow up or death from disease as the second interval (distant disease speci c survival (DDSS)) separately and combined. We compared distant recurrence lead time by detection method to DDFI and DDSS, the two component time intervals of disease speci c survival, and modeled the relative contribution of detection method to DDFI. We also assessed relative rMBC incidence by detection method.

Study Design
We conducted a retrospective cohort analysis of all rst primary stage I-III invasive BC cases age 40-74 from 1990-2016, with follow-up through 2018 for distant recurrence and vital status (n = 6603). Age 40-74 years was selected based on screening recommendations during this time period. 10,14,15 Non-surgical cases (n = 18), patients who refused recommended treatment other than surgery (n = 24), cases with unknown method of detection (n = 11) and cases with unknown cancer status at follow up (n = 139) were excluded from the analysis. In ammatory breast cancer (T4) cases were excluded (n = 125), as the diagnosis is symptom based and not detected by mammography. Patient (PtD) and mammography detected (MamD) BC was included and BC found by a medical professional from a lump or abnormality during routine physical examination were excluded (n = 295). (Fig. 1) Our institutional breast cancer registry database contains detailed information on diagnosis, pathology, staging, treatment, tumor markers, and vital status at follow up including cause-speci c death. Incident BC cases are entered at time of diagnosis into the HIPAA compliant and IRB approved registry. Patient vital and disease status including date, site and type of recurrence and date and cause of death are collected prospectively through annual updates by a certi ed cancer registrar complete through 2018 for this cohort. Follow-up status was obtained from 1) electronic chart review, 2) IRB-approved physician directed follow-up letter, 3) the institution's cancer registry, and 4) SEER Seattle-Puget Sound Registry. 16 Distant disease recurrence (rMBC) was restricted to rst presentation of distant disease excluding dates of subsequent disease progression. Hormone receptor positivity was estrogen and/or progesterone receptor positive (HR positive) and HR negative if negative for both. Self-reported race was coded white/non-white. All cases were coded to AJCC 7 classic anatomic staging across all years. 17 TNM stage 0 were excluded from the analysis as few were patient detected and distant recurrence was a rare event. Distant recurrence (rMBC) was designated if distant disease diagnosis occurred three months or more post initial diagnosis.
Breast cancer detection method was obtained by medical record review by a certi ed cancer registrar.
Mammography detected was assigned to breast cancer discovered by routine mammography in the absence of complaints or known physical ndings or as a repeat or diagnostic mammogram to verify a previous equivocal mammography nding. Patient detection was assigned if the patient presented with personally detected breast symptoms, such as a palpable lump, pain, swelling, nipple discharge, or bleeding which prompted a doctor visit.
Patients with self-detected tumors may have subsequently had a mammogram or ultrasound done but would still be categorized as a patient-detected breast cancer from rst presentation. Detection method was recorded by the physician at time of diagnosis and was only assigned if it was certain from the record.
Pearson chi-square test comparisons of categorical characteristics by detection method and mean comparisons for continuous variables were used (F statistic). Distant disease-free interval (DDFI) was time from primary BC diagnosis to distant recurrence, distant disease speci c survival (DDSS) was time from distant recurrence (rMBC) to last follow-up or death from this disease, and disease speci c survival (DSS) was total time from initial BC diagnosis to last follow-up or death from this disease. By dividing DSS into two component parts, time to distant disease recurrence (DDFI) and time from distant disease recurrence to last follow up or death from disease (DDSS), we are able to identify which portion of survival time is affected by lead time and evaluate accordingly. Kaplan-Meier estimation was used to calculate 5-year DDFI, DDSS and DSS rates (log rank tests).
Covariates signi cant by detection method were used to build the model, informed by the chi-square analysis and tested a priori using stepwise entry. The multivariable Cox proportional hazards model was used to estimate adjusted hazard ratios (HzR) and corresponding 95% con dence intervals (CI) using DDFI as the outcome. We evaluated the proportional hazards assumption by plotting ln{−ln(survival)} curves for the ordinal covariate of diagnosis year versus ln (at risk time) and on the basis of Schoenfeld residuals after tting individual Cox models.
We found no evidence suggesting substantial violation of the proportionality assumption graphically or in tests for interaction with the logarithm of survival time. 18 Effect modi cation was evident from the Cox proportional hazards analysis with stage the dominant variable in the model. Therefore, lead time analysis was strati ed by stage to compare detection method differences in survival. 19,20 All p-values were 2-sided and analyses were conducted using SPSS v.26. 21

Discussion
The majority of distant disease recurrence (68%) occurred among patient-detected cases with an incidence rate of 11% compared to 3% distant recurrence among mammography detected breast cancer cases. Mean time to distant recurrence was shorter for patient-detected BC than it was for mammography-detected BC but did not differ signi cantly when strati ed by the effect modi er stage at diagnosis which was more strongly associated with the

Strengths and Limitations
Mammography screening in the United States relies on opportunistic mammography screens based on United States Preventive Services Task Force, the American Cancer Society and other organizations recommendations unlike countries with organized screening programs. 14, 15, 22, 23 Screening is therefore predicated on self-initiation of screening mammography or prompted by a care provider or health care system. In the absence of a national screening program and as screening participation data connected to outcomes is not readily available, we tested the lead time bias hypothesis using an institutional cohort and mammography detected breast cancer as a proxy for screen detection compared to patient detected breast cancer.
Mammography screening participation rate reported in the year prior to 2012 was 57% in Washington State, very close to our observed rate of 61% mammography detection. 24 We do not have information regarding age appropriate mammography screening program participation or time interval between last non-diagnostic mammogram and breast cancer discovered by mammography. While it has been speculated that some mammography screen detected cancer would not become clinically evident in a woman's life time, to date there are no published reports of screen detected breast cancer regression or spontaneous disappearance. 25 Only invasive breast cancer stage I-III were included in the analysis as stage 0 may be interpreted as an overdiagnosis category detected by mammography and therefore not compatible with survival comparison by detection method. 26 Interpretation: comparison to other studies In a study of 233 patients diagnosed in 1988 of patients with stage 0-IV breast cancer, mammography screendetected (MSDG) breast cancer had superior prognosis primarily due to the mammography detected better prognosis, low stage breast cancer at diagnosis compared to the non-mammography screen detected group.
Superior prognosis was mainly because of the lower stage at diagnosis among the screened group as adjusting for stage removed the difference as we observed in our study. 27 In a study of screen and non-screen detected breast cancer by Lawrence et al, 1988Lawrence et al, -2004 10-year disease-speci c survival was 86% for screened and 64% in non-screened symptomatic cases. 28

Declarations
Ethics approval and consent to participate: All work on this analysis and publication was IRB approved by the Providence Institutional Review Board. As the analysis only used de-identi ed data no additional approval from patients was required and the study had exempt status.