CoChat
Sign Up
Cited by 290Artificial Intelligence in Healthcare and EducationHip and Femur FracturesRadiomics and Machine Learning in Medical Imaging

Deep learning predicts hip fracture using confounding patient and healthcare variables

Open access
Badgeley, Marcus A., Zech, John R., Oakden-Rayner, Luke, Glicksberg, Benjamin S., Liu, Manway, Gale, William, McConnell, Michael V., Percha, Bethany +2 more·2019·npj Digital Medicine
DOI 10.1038/s41746-019-0105-1PMC PMC6550136
Open this paper in CoChat — free accountOpen PDF

Abstract

Hip fractures are a leading cause of death and disability among older adults. Hip fractures are also the most commonly missed diagnosis on pelvic radiographs, and delayed diagnosis leads to higher cost and worse outcomes. Computer-aided diagnosis (CAD) algorithms have shown promise for helping radiologists detect fractures, but the image features underpinning their predictions are notoriously difficult to understand. In this study, we trained deep-learning models on 17,587 radiographs to classify fracture, 5 patient traits, and 14 hospital process variables. All 20 variables could be individually predicted from a radiograph, with the best performances on scanner model (AUC = 1.00), scanner brand (AUC = 0.98), and whether the order was marked “priority” (AUC = 0.79). Fracture was predicted moderately well from the image (AUC = 0.78) and better when combining image features with patient data (AUC = 0.86, DeLong paired AUC comparison, p = 2e-9) or patient data plus hospital process features (AUC = 0.91, p = 1e-21). Fracture prediction on a test set that balanced fracture risk across patient variables was significantly lower than a random test set (AUC = 0.67, DeLong unpaired AUC comparison, p = 0.003); and on a test set with fracture risk balanced across patient and hospital process variables, the model performed randomly (AUC = 0.52, 95% CI 0.46–0.58), indicating that these variables were the main source of the model’s fracture predictions. A single model that directly combines image features, patient, and hospital process data outperforms a Naive Bayes ensemble of an image-only model prediction, patient, and hospital process data. If CAD algorithms are inexplicably leveraging patient and process variables in their predictions, it is unclear how radiologists should interpret their predictions in the context of other known patient data. Further research is needed to illuminate deep-learning decision processes so that computers and clinicians can effectively cooperate.

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1038/s41746-019-0105-1

Cites

7 of 7 references verified0 citation errors · 0 not checkable

Statistics

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

  • Large performance disparities across users can indicate that models are exploiting biases or spurious correlations in the data.

    “These large performance disparities across users are practical concerns in consumer-facing applications, and they can also indicate that models are exploiting biases or spurious correlations in the data (Badgeley et al., 2019; Geva et al., 2019).”

    From WILDS: A Benchmark of in-the-Wild Distribution Shifts (2020) · cites this paper as Badgeley et al., 2019

    The Results section (‘Evaluating CNNs on matched patient populations’; page unavailable) reports that fracture-detection performance fell as patient and hospital-process variables were matched, and the Discussion says this revealed dependence on confounding variables. This supports the idea that performance differences can expose spurious associations, but the paper compares matched patient cohorts, not performance disparities across users of a consumer-facing application. (LLM, conf=0.83)

    Experimental — accuracy not yet measured

  • Large performance disparities across users are practical concerns in consumer-facing applications and can indicate that models are exploiting biases or spurious correlations in the data.

    “These large performance disparities across users are practical concerns in consumer-facing applications, and they can also indicate that models are exploiting biases or spurious correlations in the data (Badgeley et al., 2019; Geva et al., 2019).”

    From WILDS: A Benchmark of in-the-Wild Distribution Shifts (2020) · cites this paper as Badgeley et al., 2019

    The Introduction (page unavailable) says models can learn unintended patterns and confounding variables. The Results, under “Evaluating CNNs on matched patient populations” (page unavailable), report substantially lower fracture-detection performance after matching for patient and hospital-process variables. This supports the concern about spurious correlations, but the paper does not address performance disparities across users or practical concerns in consumer-facing applications. (LLM, conf=0.85)

    Experimental — accuracy not yet measured

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Cites

Flagged first, then by citations

Research works printed but not linked: 40

  1. [1]Johnell O Kanis JA An estimate of the worldwide prevalence, mortality and disability associated with hip fracture Osteoporos. Int. 2004 15 897 902 10.1007/s00198-004-1627-0 15490120 · DOI
  2. [2]Haentjens P Meta-analysis: excess mortality after hip fracture among older women and men Ann. Intern. Med. 2010 152 380 390 10.7326/0003-4819-152-6-201003160-00008 20231569 PMC3010729 · DOI
  3. [4]Kirby MW Spritzer C Radiographic detection of hip and pelvic fractures in the emergency department Am. J. Roentgenol. 2010 194 1054 1060 10.2214/AJR.09.3295 20308510 · DOI
  4. [5]Cannon J Silvestri S Munro M Imaging choices in occult hip fracture J. Emerg. Med. 2009 37 144 152 10.1016/j.jemermed.2007.12.039 18963720 · DOI
  5. [6]Donald JJ Barnard SA Common patterns in 558 diagnostic radiology errors J. Med. Imaging Radiat. Oncol. 2012 56 173 178 10.1111/j.1754-9485.2012.02348.x 22498190 · DOI
  6. [7]Donnelley M Knowles G Hearn TA CAD system for long-bone segmentation and fracture detection Image Signal Process., Lect. notes 2008 5099 153 162 10.1007/978-3-540-69905-7_18 · DOI
  7. [8]Kazi A Automatic classification of proximal femur fractures based on attention models Mach. Learn. Med. Imaging, Lect. notes 2017 10541 70 78 10.1007/978-3-319-67389-9_9 · DOI
  8. [9]Chai HY Wee LK Swee TT Hussain S Gray-level co-occurrence matrix bone fracture detection WSEAS Trans. Syst. 2011 10 7 16
  9. [10]Kasai S Li F Shiraishi J Doi K Usefulness of computer-aided diagnosis schemes for vertebral fractures and lung nodules on chest radiographs AJR Am. J. Roentgenol. 2008 191 260 265 10.2214/AJR.07.3091 18562756 · DOI
  10. [11]Kim DH MacKinnon T Artificial intelligence in fracture detection: transfer learning from deep convolutional neural networks Clin. Radiol. 2018 73 439 445 10.1016/j.crad.2017.11.015 29269036 · DOI
  11. [12]Chung SW Automated detection and classification of the proximal humerus fracture by using deep learning algorithm Acta Orthop. 2018 89 468 473 10.1080/17453674.2018.1453714 29577791 PMC6066766 · DOI
  12. [13]Olczak J Artificial intelligence for analyzing orthopedic trauma radiographs Acta Orthop. 2017 88 581 586 10.1080/17453674.2017.1344459 28681679 PMC5694800 · DOI
  13. [14]Gale, W., Oakden-Rayner, L., Carneiro, G., Bradley, A. P. & Palmer, L. J. Detecting hip fractures with radiologist-level performance using deep neural networks. Preprint at https://arxiv.org/abs/1711.06504 (2017). · arXiv
  14. [15]Buolamwini J Gebru T Gender shades: intersectional accuracy disparities in commercial gender classification Proc. 1st Conf. Fairness, Account., Transpar., PMLR 2018 81 77 91
  15. [16]Kuehlkamp, A., Becker, B. & Bowyer, K. In: O’Conner, L. (ed.) Gender-from-Iris or Gender-from-Mascara? In 2017 IEEE Winter Conference on Applications of Computer Vision (WACV). 1151–1159, IEEE Computer Society: Los Alamitos, California, 2017). https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=7926586.
  16. [17]Caliskan A Bryson JJ Narayanan A Semantics derived automatically from language corpora contain human-like biases Science 2017 356 183 186 10.1126/science.aal4230 28408601 · DOI · Paper page
  17. [18]Agniel D Kohane IS Weber GM Biases in electronic health record data due to processes within the healthcare system: retrospective observational study BMJ 2018 361 k1479 10.1136/bmj.k1479 29712648 PMC5925441 · DOI · Paper page
  18. [19]Grimes DA Schulz KF Bias and causal associations in observational research Lancet 2002 356 9302 11812579
  19. [20]Poplin R Prediction of cardiovascular risk factors from retinal fundus photographs via deep learning Nat. Biomed. Eng. 2018 2 158 164 10.1038/s41551-018-0195-0 31015713 · DOI · Paper page
  20. [22]Madani A Arnaout R Mofrad M Arnaout R Fast and accurate view classification of echocardiograms using deep learning npj Digit. Med. 2018 1 6 10.1038/s41746-017-0013-1 30828647 PMC6395045 · DOI · Paper page
  21. [23]Yao, L. et al. Learning to diagnose from scratch by exploiting dependencies among labels. Preprint at https://arxiv.org/abs/1710.10501 (2017). · arXiv
  22. [24]Kanis JA Assessment of fracture risk Osteoporos. Int. 2005 16 581 589 10.1007/s00198-004-1780-5 15616758 · DOI
  23. [25]Pearce N Analysis of matched case-control studies BMJ 2016 352 i969 10.1136/bmj.i969 26916049 PMC4770817 · DOI
  24. [26]Brestel, C., Shadmi, R., Tamir, I., Cohen-Sfaty, M. & Elnekave, E. RadBot-CXR: classification of four clinical finding categories in chest X-ray using deep learning. Preprint at https://openreview.net/pdf?id=B1tMhcIDM (2018).
  25. [27]Terp S Enforcement of the emergency medical treatment and labor act, 2005 to 2014 Ann. Emerg. Med. 2017 69 155 162 10.1016/j.annemergmed.2016.05.021 27496388 PMC7176068 · DOI
  26. [28]Titano JJ Automated deep neural network surveillance of cranial images for acute neurologic events Nat. Med. 2018 24 1337 1341 10.1038/s41591-018-0147-y 30104767 · DOI · Paper page
  27. [29]Rajpurkar, P. et al. CheXNet: radiologist-level pneumonia detection on chest X-rays with deep learning. Preprint at https://arxiv.org/abs/1712.06957 (2017). · arXiv
  28. [30]Rajpurkar, P. et al. MURA dataset: towards radiologist-level abnormality detection in musculoskeletal radiographs. arXiv [physics.med-ph] (2017).
  29. [31]Islam, M. T., Aowal, M. A., Minhaz, A. T. & Ashraf, K. Abnormality detection and localization in chest X-rays using deep convolutional neural networks. Preprint at https://arxiv.org/abs/1705.09850 (2017). · arXiv
  30. [32]Tomita N Cheung YY Hassanpour S Deep neural networks for automatic detection of osteoporotic vertebral fractures on CT scans Comput. Biol. Med. 2018 98 8 15 10.1016/j.compbiomed.2018.05.011 29758455 · DOI
  31. [34]Shin, H.-C. et al. In: O’Conner, L. (ed.) Learning to read chest X-rays: recurrent neural cascade model for automated image annotation In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2497–2506, IEEE Computer Society: Los Alamitos, California, 2016). https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=7780332.
  32. [35]Roth, H. R. et al. In: Navab, N., Hornegger, J., Wells, W. & Frangi, A. (eds.), DeepOrgan: multi-level deep convolutional networks for automated pancreas segmentation In Medical Image Computing and Computer-Assisted Intervention-MICCAI. 564–556 Springer, Cham: Switzerland, 2015). https://link.springer.com/book/10.1007%2F978-3-319-24553-9#. · DOI
  33. [36]Roth, H. R. et al. Deep convolutional networks for automated detection of posterior-element fractures on spine CT. Preprint at https://arxiv.org/abs/1602.00020 (2016). · arXiv
  34. [37]Abràmoff MD Lavin PT Birch M Shah N Folk JC Pivotal trial of an autonomous AI-based diagnostic system for detection of diabetic retinopathy in primary care offices npj Digit. Med. 2018 1 39 10.1038/s41746-018-0040-6 PMC6550188 31304320 · DOI · Paper page
  35. [38]DeFauw J Clinically applicable deep learning for diagnosis and referral in retinal disease Nat. Med. 2018 24 1342 1350 10.1038/s41591-018-0107-6 30104768 · DOI · Paper page
  36. [39]Zhang J Patient2Vec: a personalized Interpretable Deep Representation of the Longitudinal Electronic Health Record IEEE Access 2018 6 65333 65346 10.1109/ACCESS.2018.2875677 · DOI
  37. [43]Hoffman GE Schadt E E. variancePartition: interpreting drivers of variation in complex gene expression studies BMC Bioinforma. 2016 17 483 10.1186/s12859-016-1323-z PMC5123296 27884101 · DOI · Paper page
  38. [45]Bousmalis, K., Trigeorgis, G., Silberman, N., Krishnan, D. & Erhan, D. Domain Separation Networks. Preprint at https://arxiv.org/abs/1608.06019 (2016). · arXiv
  39. [47]Szegedy, C., Vanhoucke, V., Ioffe, S., Shlens, J. & Wojna, Z. Rethinking the Inception Architecture for Computer Vision. Preprint at https://arxiv.org/abs/1512.00567 (2015). · arXiv
  40. [48]Russakovsky O ImageNet large scale visual recognition challenge Int J. Comput Vis. 2015 115 211 10.1007/s11263-015-0816-y · DOI · Paper page

Cited by

Sorted by citations

Authors
Badgeley, M. A., Zech, J. R., Oakden-Rayner, L., Glicksberg, B. S., Liu, M., Gale, W., McConnell, M. V., Percha, B., Snyder, T. M., & Dudley, J. T.
Affiliations
  • Verily Life Sciences LLC (United States) (US)
  • California Pacific Medical Center (US)
  • University of California, San Francisco (US)
  • The University of Adelaide (AU)
  • Stanford Medicine (US)
  • Icahn School of Medicine at Mount Sinai (US)
Corresponding author
Dudley, Joel T.
ORCID
Venue
npj Digital Medicine
Publisher
Springer Science and Business Media LLC
Published
2019-04-30
Volume
2
Issue
1
Pages
31-31
ISSN
2398-6352 (ISSN-L)
DOI
10.1038/s41746-019-0105-1
PMC
PMC6550136
OpenAlex
W2900210626
Cited by
290
References
7
Citations by year
2026: 35 · 2025: 34 · 2024: 37 · 2023: 30 · 2022: 52 · 2021: 53 · 2020: 40 · 2019: 8 · 2016: 1
FWCI
12.13
Type
journal article
Language
en
Open access
Licence
Locations
  • npj Digital Medicine (publishedVersion, cc-by)
  • Europe PMC (PubMed Central) (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • arXiv (Cornell University) (submittedVersion)
  • arXiv (Cornell University)
Funders
U.S. Department of Health & Human Services | National Institutes of Health (UL1TR001433-01)
Topics
  • Health Sciences › Medicine › Health Informatics › Artificial Intelligence in Healthcare and Education
  • Health Sciences › Medicine › Surgery › Hip and Femur Fractures
  • Health Sciences › Medicine › Radiology, Nuclear Medicine and Imaging › Radiomics and Machine Learning in Medical Imaging
Keywords
hip fracture, computer-aided diagnosis, pelvic radiographs, confounding variables, patient characteristics, deep learning, explainable AI, fracture risk prediction, radiographic interpretation

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review