The Predictive Value of Specific Mammographic Findings in Breast Cancer Detection

1
San Francisco Mammography Registry

• The Predictive Value of Specific Mammographic
  Findings in Breast Cancer Detection

Aruna Venkatesan1, BSE Philip Chu2, MS Karla
Kerlikowske3,4, MD Rebecca Smith-Bindman2,3, MD
1 School of Medicine 2 Department of
Radiology 3 Department of Epidemiology and
Biostatistics 4 Department of Medicine
2
Background

• Mammography is standard of care for breast cancer
  detection, yet it is imprecise with a high false
  positive rate
• Specific findings that raise suspicion of cancer
  have not been adequately studied in large,
  diverse populations
• Mammography accuracy may be improved with better
  knowledge of the predictive values of specific
  findings

3
Research Questions

• Among mammograms with a positive result
• (1) What is the distribution of specific
  mammographic findings among women with and
  without cancer?
• (2) What is the yield of cancer, or positive
  predictive value (PPV), for specific findings?
• (3) How do the PPVs of each finding vary by
  radiology reader and patient factors such as age
  or family history?

4
Methods

• Study design and population
• Prospective data collection by San Francisco
  Mammography Registry from January 1998 to
  December 2002
• 10,520 exams among 8,750 women with recorded
  mammographic findings from 7 sites in the SFMR
• Excluded women with history of breast cancer,
  lumpectomy, mastectomy, radiation therapy, or
  breast reduction

5
Methods

• Definitions
• Type of Exam Screening or Diagnostic
• Positive Result BI-RADS Assessment 0, 3, 4, or
  5
• Specific Findings
• Mass
• Calcification
• Focal Asymmetry
• Architectural Distortion
• Breast Cancer DCIS or Invasive Cancer diagnosed
  within 14 months of mammogram

6
Methods

• SFMR

1. Performed Linkage between SFMR and SEER Cancer
Registry
Cancers Diagnosed through 2003
SEER
2. Designed Tables and Data Analysis 3. Performed
Analysis using STATA and SAS Software
7
Results

• Demographics
• Age
• Women with cancer were disproportionately older
  than the general study sample
• Race and Ethnicity
• White women represented more cancers compared to
  women of other races and ethnicities
• Cancer
• 17 of study participants were diagnosed with
  cancer
• 82.7 Invasive Cancer
• 17.3 Ductal Carcinoma in Situ (DCIS)

8
Distribution and Yield of Cancer by Finding
False Positives Invasive Cancer
DCIS PPV()
PPV() Screening Mammograms Mass 35.0 48.7 9.6
3.1 0.2 Calcification 32.1 28.5 5.9 95.8 6.9
Focal Asymm 26.4 13.0 3.6 1.0 1.1 Arch
Distortion 6.5 9.7 10.4 0.0 0.0 Diagnostic
Mammograms Mass 70.0 72.4 19.1 16.1 0.8 Calci
fication 25.4 19.5 13.4 82.2 11.0 Focal
Asymm 3.8 2.6 13.5 1.3 1.3 Arch
Distortion 0.8 5.4 59.8 0.4 0.9
9
Reader Variation in PPV for Mass vs. Focal
Asymmetry
10
Conclusions

• Masses predict invasive cancer
• Calcifications alone predict both DCIS and
  invasive cancer
• Focal Asymmetries have a low PPV for invasive
  cancer
• Architectural distortion has a low prevalence but
  a high PPV for invasive cancer

11
Conclusions

• Focal asymmetries demonstrate
• High inter-reader variability in PPV
• A low PPV of 3.7 in screening exams, while
  comprising about 25 of all screening mammograms
• Comprise 11.6 of abnormal findings but only 4.0
  of cancers detected
• RECOMMENDATION Reconsider whether focal
  asymmetry is a useful finding when interpreting
  screening mammograms, especially in women less
  than 70 years of age

12
Acknowledgments

• Dr. Rebecca Smith-Bindman and Philip Chu For
  their wisdom, guidance, and mentorship
• Dr. Karla Kerlikowske, Principal Investigator of
  SFMR
• For her ideas and input throughout the process
• Dr. Edward Sickles, Retired Chief of Breast
  Imaging
• For his clinical expertise
• San Francisco Mammography Registry
• Senior Statistician Michael Hofmann
• Deans Summer Research Fellowship