Perspective | Published:


Machine-learning-based patient-specific prediction models for knee osteoarthritis

Nature Reviews Rheumatologyvolume 15pages4960 (2019) | Download Citation


Osteoarthritis (OA) is an extremely common musculoskeletal disease. However, current guidelines are not well suited for diagnosing patients in the early stages of disease and do not discriminate patients for whom the disease might progress rapidly. The most important hurdle in OA management is identifying and classifying patients who will benefit most from treatment. Further efforts are needed in patient subgrouping and developing prediction models. Conventional statistical modelling approaches exist; however, these models are limited in the amount of information they can adequately process. Comprehensive patient-specific prediction models need to be developed. Approaches such as data mining and machine learning should aid in the development of such models. Although a challenging task, technology is now available that should enable subgrouping of patients with OA and lead to improved clinical decision-making and precision medicine.

Access optionsAccess options

Rent or Buy article

Get time limited or full article access on ReadCube.


All prices are NET prices.

Additional information

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.




Automatic Image Registration:




  1. 1.

    Arden, N. & Cooper, Osteoarthritis Handbook (Taylor & Francis, London, 2006).

  2. 2.

    McGuire, D. A., Carter, T. R. & Shelton, W. R. Complex knee reconstruction: osteotomies, ligament reconstruction, transplants, and cartilage treatment options. Arthroscopy 18, 90–103 (2002).

  3. 3.

    Cooper, C. & Arden, N. K. Excess mortality in osteoarthritis. BMJ 342, d1407 (2011).

  4. 4.

    Hochberg, M. C. Mortality in osteoarthritis. Clin. Exp. Rheumatol 26, S120–S124 (2008).

  5. 5.

    Bitton, R. The economic burden of osteoarthritis. Am. J. Manag. Care 15, S230–S235 (2009).

  6. 6.

    Prieto-Alhambra, D. et al. Incidence and risk factors for clinically diagnosed knee, hip and hand osteoarthritis: influences of age, gender and osteoarthritis affecting other joints. Ann. Rheum. Dis. 73, 1659–1664 (2014).

  7. 7.

    Martel-Pelletier, J. et al. Osteoarthritis. Nat. Rev. Dis. Primers 2, 16072 (2016).

  8. 8.

    Blagojevic, M., Jinks, C., Jeffery, A. & Jordan, K. P. Risk factors for onset of osteoarthritis of the knee in older adults: a systematic review and meta-analysis. Osteoarthritis Cartilage 18, 24–33 (2010).

  9. 9.

    Zhang, W. Risk factors of knee osteoarthritis — excellent evidence but little has been done. Osteoarthritis Cartilage 18, 1–2 (2010).

  10. 10.

    McWilliams, D. F., Leeb, B. F., Muthuri, S. G., Doherty, M. & Zhang, W. Occupational risk factors for osteoarthritis of the knee: a meta-analysis. Osteoarthritis Cartilage 19, 829–839 (2011).

  11. 11.

    Raynauld, J. P. et al. Long term evaluation of disease progression through the quantitative magnetic resonance imaging of symptomatic knee osteoarthritis patients: correlation with clinical symptoms and radiographic changes. Arthritis Res. Ther. 8, R21 (2006).

  12. 12.

    Solomon, D. H. et al. The comparative safety of analgesics in older adults with arthritis. Arch. Intern. Med. 170, 1968–1978 (2010).

  13. 13.

    Marx, V. Biology: the big challenges of big data. Nature 498, 255–260 (2013).

  14. 14.

    Dolinski, K. & Troyanskaya, O. G. Implications of big data for cell biology. Mol. Biol. Cell 26, 2575–2578 (2015).

  15. 15.

    Cintolo-Gonzalez, J. A. et al. Breast cancer risk models: a comprehensive overview of existing models, validation, and clinical applications. Breast Cancer Res. Treat. 164, 263–284 (2017).

  16. 16.

    Cosma, G., Brown, D., Archer, M., Khan, M. & Pockley, A. G. A survey on computational intelligence approaches for predictive modeling in prostate cancer. Expert Syst. Appl. 70, 1–19 (2017).

  17. 17.

    Fast and Secure protocol — FASP (Aspera, Inc., Emeryville, CA, USA).

  18. 18.

    Zhang, W. et al. Nottingham knee osteoarthritis risk prediction models. Ann. Rheum. Dis. 70, 1599–1604 (2011).

  19. 19.

    Losina, E., Klara, K., Michl, G. L., Collins, J. E. & Katz, J. N. Development and feasibility of a personalized, interactive risk calculator for knee osteoarthritis. BMC Musculoskelet. Disord. 16, 312 (2015).

  20. 20.

    Watt, E. W. & Bui, A. A. Evaluation of a dynamic Bayesian belief network to predict osteoarthritic knee pain using data from the osteoarthritis initiative. AMIA Annu. Symp. Proc. 2008, 788–792 (2008).

  21. 21.

    Yoo, T. K., Kim, D. W., Choi, S. B., Oh, E. & Park, J. S. Simple scoring system and artificial neural network for knee osteoarthritis risk prediction: a cross-sectional study. PLoS ONE 11, e0148724 (2016).

  22. 22.

    Lazzarini, N. et al. A machine learning approach for the identification of new biomarkers for knee osteoarthritis development in overweight and obese women. Osteoarthritis Cartilage 25, 2014–2021 (2017).

  23. 23.

    Schett, G. et al. Vascular cell adhesion molecule 1 as a predictor of severe osteoarthritis of the hip and knee joints. Arthritis Rheum. 60, 2381–2389 (2009).

  24. 24.

    Schett, G., Zwerina, J., Axmann, R., Willeit, J. & Stefan, K. Risk prediction for severe osteoarthritis. Ann. Rheum. Dis. 69, 1573–1574 (2010).

  25. 25.

    Berthiaume, M. J. et al. Meniscal tear and extrusion are strongly associated with the progression of knee osteoarthritis as assessed by quantitative magnetic resonance imaging. Ann. Rheum. Dis. 64, 556–563 (2005).

  26. 26.

    Raynauld, J. P. et al. Correlation between bone lesion changes and cartilage volume loss in patients with osteoarthritis of the knee as assessed by quantitative magnetic resonance imaging over a 24-month period. Ann. Rheum. Dis. 67, 683–688 (2008).

  27. 27.

    Tanamas, S. K. et al. Bone marrow lesions in people with knee osteoarthritis predict progression of disease and joint replacement: a longitudinal study. Rheumatology 49, 2413–2419 (2010).

  28. 28.

    Raynauld, J. P. et al. Risk factors predictive of joint replacement in a 2-year multicentre clinical trial in knee osteoarthritis using MRI: results from over 6 years of observation. Ann. Rheum. Dis. 70, 1382–1388 (2011).

  29. 29.

    Pelletier, J. P. et al. What is the predictive value of MRI for the occurrence of knee replacement surgery in knee osteoarthritis? Ann. Rheum. Dis. 72, 1594–1604 (2013).

  30. 30.

    Neogi, T. et al. Magnetic resonance imaging-based three-dimensional bone shape of the knee predicts onset of knee osteoarthritis: data from the osteoarthritis initiative. Arthritis Rheum. 65, 2048–2058 (2013).

  31. 31.

    Raynauld, J. P. et al. Bone curvature changes can predict the impact of treatment on cartilage volume loss in knee osteoarthritis: data from a 2-year clinical trial. Rheumatology 56, 989–998 (2017).

  32. 32.

    Fan, J., Han, F. & Liu, H. Challenges of big data analysis. Natl Sci. Rev. 1, 293–314 (2014).

  33. 33.

    Haixiang, G. et al. Learning from class-imbalanced data: review of methods and applications. Expert Syst. Appl. 73, 220–239 (2017).

  34. 34.

    Fu, X., Wang, L., Chua, K. S. & Chu, F. Training RBF Neural Networks on Unbalanced Data. Proc. 9th Int. Conf. Neural Inform. Processing (ICONIP’02) 2, 1016–1020 (2002).

  35. 35.

    Wasikowski, M. & Chen, X. W. Combating the small sample class imbalance problem using feature selection. IEEE Trans. Knowl. Data Eng. 22, 1388–1400 (2010).

  36. 36.

    Khalilia, M., Chakraborty, S. & Popescu, M. Predicting disease risks from highly imbalanced data using random forest. BMC Med. Inform. Decis. Mak. 11, 51 (2011).

  37. 37.

    Wang, K. J., Makond, B. & Wang, K. M. An improved survivability prognosis of breast cancer by using sampling and feature selection technique to solve imbalanced patient classification data. BMC Med. Inform. Decis. Mak. 13, 124 (2013).

  38. 38.

    Ozcift, A. Random forests ensemble classifier trained with data resampling strategy to improve cardiac arrhythmia diagnosis. Comput. Biol. Med. 41, 265–271 (2011).

  39. 39.

    van Buuren, S. & Groothuis-Oudshoorn, K. MICE: multivariate imputation by chained equations in R. J. Stat. Softw. 45, 1–68 (2011).

  40. 40.

    IBM SPSS Statistics for Windows, version 25.0, released 2017 (IBM Corp., Armonk, NY, USA).

  41. 41.

    SAS/STAT® version 14.1 (SAS Institute Inc., Cary, NC, USA).

  42. 42.

    STATA Statistical Software, release 15, 2017 (StataCorp LLC, College Station, TX, USA).

  43. 43.

    Frank, E., Hall, M. A. & Witten, I. H. The WEKA workbench: online appendix for data mining: practical machine learning tools and techniques. UoW (2016).

  44. 44.

    Zhang, Z. Missing data imputation: focusing on single imputation. Ann. Transl Med. 4, 9 (2016).

  45. 45.

    Verborgh, R. & De Wilde, M. Using OpenRefine (Packt Publishing Ltd., Burmingham, UK, 2013).

  46. 46.

    Trifacta. Data wrangling tools & software. Trifacta (2018).

  47. 47.

    Paxata, Inc. Self-service data preparation for data analytics. Paxata (2018).

  48. 48.

    Baruti, R. (ed.) Learning Alteryx: A Beginner’s Guide to Using Alteryx for Self-Service Analytics and Business Intelligence (Packt Publishing Ltd., Birmingham, UK, 2017).

  49. 49.

    McKinney, W. pandas: a foundational python library for data analysis and statistics. DLR (2011).

  50. 50.

    OBiBa. Open source software for epidemiology. OBiBa (2018).

  51. 51.

    Optimus Company. Data cleansing and exploration made simple. Optimus (2018).

  52. 52.

    Griffith, L. E. et al. Statistical approaches to harmonize data on cognitive measures in systematic reviews are rarely reported. J. Clin. Epidemiol. 68, 154–162 (2015).

  53. 53.

    Royston, P., Parmar, M. K. & Sylvester, R. Construction and validation of a prognostic model across several studies, with an application in superficial bladder cancer. Stat. Med. 23, 907–926 (2004).

  54. 54.

    Doiron, D. et al. Data harmonization and federated analysis of population-based studies: the BioSHaRE project. Emerg. Themes Epidemiol. 10, 12 (2013).

  55. 55.

    Doiron, D., Raina, P., Ferretti, V., L’Heureux, F. & Fortier, I. Facilitating collaborative research: implementing a platform supporting data harmonization and pooling. Nor. Epidemiol. 21, 221–224 (2012).

  56. 56.

    Alba, A. C. et al. Discrimination and calibration of clinical prediction models: users’ guides to the medical literature. JAMA 318, 1377–1384 (2017).

  57. 57.

    Steyerberg, E. W. & Harrell, F. E. Jr. Prediction models need appropriate internal, internal-external, and external validation. J. Clin. Epidemiol. 69, 245–247 (2016).

  58. 58.

    Siontis, G. C., Tzoulaki, I., Castaldi, P. J. & Ioannidis, J. P. External validation of new risk prediction models is infrequent and reveals worse prognostic discrimination. J. Clin. Epidemiol. 68, 25–34 (2015).

  59. 59.

    Tugwell, P. & Knottnerus, J. A. Clinical prediction models are not being validated. J. Clin. Epidemiol. 68, 1–2 (2015).

  60. 60.

    Tugwell, P. & Knottnerus, J. A. Transferability/generalizability deserves more attention in ‘retest’ studies in diagnosis and prognosis. J. Clin. Epidemiol. 68, 235–236 (2015).

  61. 61.

    Debray, T. P., Moons, K. G., Ahmed, I., Koffijberg, H. & Riley, R. D. A framework for developing, implementing, and evaluating clinical prediction models in an individual participant data meta-analysis. Stat. Med. 32, 3158–3180 (2013).

  62. 62.

    Debray, T. P. et al. A new framework to enhance the interpretation of external validation studies of clinical prediction models. J. Clin. Epidemiol. 68, 279–289 (2015).

  63. 63.

    Steyerberg, E. Clinical Prediction Models: A Practical Approach to Development, Validation, and Updating (Springer New York, 2010).

  64. 64.

    Papageorgiou, E. I., Subramanian, J., Karmegam, A. & Papandrianos, N. A risk management model for familial breast cancer: a new application using Fuzzy Cognitive Map method. Comput. Methods Programs Biomed. 122, 123–135 (2015).

  65. 65.

    Froelich, W., Papageorgiou, E. I., Samarinas, M. & Skriapas, K. Application of evolutionary fuzzy cognitive maps to the long-term prediction of prostate cancer. Appl. Soft Comput. 12, 3810–3817 (2012).

  66. 66.

    Takahashi, H. et al. Prediction model for knee osteoarthritis based on genetic and clinical information. Arthritis Res. Ther. 12, R187 (2010).

  67. 67.

    Kerkhof, H. J. et al. Prediction model for knee osteoarthritis incidence, including clinical, genetic and biochemical risk factors. Ann. Rheum. Dis. 73, 2116–2121 (2014).

  68. 68.

    Kinds, M. B. et al. Evaluation of separate quantitative radiographic features adds to the prediction of incident radiographic osteoarthritis in individuals with recent onset of knee pain: 5-year follow-up in the CHECK cohort. Osteoarthritis Cartilage 20, 548–556 (2012).

  69. 69.

    Swan, A. L. et al. A machine learning heuristic to identify biologically relevant and minimal biomarker panels from omics data. BMC Genomics 16, S2 (2015).

  70. 70.

    Ashinsky, B. G. et al. Predicting early symptomatic osteoarthritis in the human knee using machine learning classification of magnetic resonance images from the osteoarthritis initiative. J. Orthop. Res. 35, 2243–2250 (2017).

  71. 71.

    Long, M. J., Papi, E., Duffell, L. D. & McGregor, A. H. Predicting knee osteoarthritis risk in injured populations. Clin. Biomech. 47, 87–95 (2017).

  72. 72.

    Minciullo, L., Bromiley, P. A., Felson, D. T. & Cootes, T. F. Indecisive trees for classification and prediction of knee osteoarthritis. 8th Int. Workshop MLMI 2017 MICCAI 2017 Proc. 10541, 283–290 (2017).

  73. 73.

    Jamshidi, A., Ait-kadi, D., Ruiz, A. & Rebaiaia, M. L. Dynamic risk assessment of complex systems using FCM. Int. J. Prod. Res. 56, 1070–1088 (2017).

  74. 74.

    Meher, S. K. & Pal, S. K. Rough-wavelet granular space and classification of multispectral remote sensing image. Appl. Soft Comput. 11, 5662–5673 (2011).

  75. 75.

    Shannon, P. et al. Cytoscape: a software environment for integrated models of biomolecular interaction networks. Genome Res. 13, 2498–2504 (2003).

  76. 76.

    Hastie, T., Tibshirani, R. & Wainwright, M. Statistical Learning with Sparsity: The Lasso and Generalizations (Chapman and Hall/CRC, 2015).

  77. 77.

    Meinshausen, N. & Bühlmann, P. High-dimensional graphs and variable selection with the lasso. Ann. Stat. 34, 1436–1462 (2006).

  78. 78.

    Huang, J., Ma, S. & Zhang, C. H. Adaptive Lasso for sparse high-dimensional regression models. Stat. Sin. 18, 1603–1618 (2008).

  79. 79.

    Simon, N., Friedman, J., Hastie, T. & Tibshirani, R. A. Sparse-group lasso. J. Comput. Graph. Stat. 22, 231–245 (2013).

  80. 80.

    Friedman, J. et al. Package ‘glmnet’. The Comprehensive R Archive Network (2018).

  81. 81.

    Alakwaa, F. M., Chaudhary, K. & Garmire, L. X. Deep learning accurately predicts estrogen receptor status in breast cancer metabolomics data. J. Proteome Res. 17, 337–347 (2018).

  82. 82.

    Nezhad, M. Z., Zhu, D., Li, X., Yang, K. & Levy, P. SAFS: a deep feature selection approach for precision medicine. Preprint at arXiv (2017).

  83. 83.

    Li, Y., Chen, C. Y. & Wasserman, W. W. Deep feature selection: theory and application to identify enhancers and promoters. J. Comput. Biol. 23, 322–336 (2016).

  84. 84.

    Christ, M., Braun, N., Neuffer, J. & Kempa-Liehr, W. A. Time series feature extraction on basis of scalable hypothesis tests (tsfresh — a Python package). Neurocomputing 307, 72–77 (2018).

  85. 85.

    Fulcher, B. D. & Jones, N. S. hctsa: a computational framework for automated time-series phenotyping using massive feature extraction. Cell Syst. 5, 527–531 (2017).

Download references


The authors thank V. Wallis for her assistance with manuscript preparation.

Author information


  1. Osteoarthritis Research Unit, University of Montreal Hospital Research Centre (CRCHUM), Montreal, Quebec, Canada

    • Afshin Jamshidi
    • , Jean-Pierre Pelletier
    •  & Johanne Martel-Pelletier


  1. Search for Afshin Jamshidi in:

  2. Search for Jean-Pierre Pelletier in:

  3. Search for Johanne Martel-Pelletier in:


J.M.-P. and A.J. researched data for the article. All authors wrote the article, made substantial contribution to discussions of the content and reviewed and/or edited the manuscript before submission.

Competing interests

The authors declare no competing interests.

Corresponding author

Correspondence to Johanne Martel-Pelletier.

Supplementary Information

GlossaryGlossary terms

Artificial intelligence

The process of creating systems that can learn from experience and adjust to new inputs in order to perform human-like tasks. Machine-learning is a fundamental concept of artificial intelligence.


Calibration measurements represent the level of accuracy of a model in estimating the absolute risk (that is, the agreement between the observed and predicted risk). Poorly calibrated models will underestimate or overestimate the outcome of interest.

Classification models

In statistics and machine-learning, classification is the process of identifying the category of a new observation on the basis of a training set of data containing observations for which the category (outcome value) is known. In the field of osteoarthritis, an example could be classification of patients into slow progressors and fast progressors on the basis of several input variables.


A subfield of machine-learning that is based on advanced artificial neural networks; this field has enabled doctors in different fields of medicine to obtain a precise 3D understanding of 2D images.


Discrimination measurements identify to what extent a model discriminates items of different classes (for example, individuals with disease and without disease). For binary outcomes, the receiver operating characteristic curve or C-statistic could be applied for discrimination measurement.

Feature selection

Feature selection refers to the process of obtaining a subset of variables from an original set of variables according to certain feature selection criteria. The feature selection step precedes the learning step of a prediction model and good feature selection results can improve the learning accuracy, reduce learning time and simplify learning results.


Refers to the accuracy with which a prediction model developed from one study population can be used for the population at large.


In machine-learning and statistics, imputation is the process of replacing missing data with substituted values to avoid bias or inaccuracies in the results.


Model interpretability describes the ability of the user to understand the model, which includes understanding the relationships between the input and outcome variables (for example, knowing how the selected input variables contribute to the outcome variable).

Regression models

Regression is the process of identifying the value of a new observation on the basis of a training set of data containing observations for which the category (outcome value) is known. In the field of osteoarthritis, an example could be predicting the probability of disease.

Semi-supervised learning

Semi-supervised learning is typically when only a small amount of data are labelled (that is, have both input and output variables) and a large amount are unlabelled (that is, have only input data); this method falls between unsupervised learning and supervised learning.

Supervised learning

Supervised learning is where you have input variables (x) and an output variable (y) and use an algorithm to learn the mapping function from the input to the output y = f(x).


The training for machine learning involves providing a machine-learning algorithm with training data (input and outcome variables) to learn from. The learning algorithm finds patterns in the training data such that the input parameters correspond to the target. Machine-learning models are applied to do predictions on new data for which the outcome value is not known (for example, to determine to which class the new observation belongs).

Unsupervised learning

In unsupervised learning, only input data (x) exist and there are no corresponding output variables. The goal for unsupervised learning is to model the underlying structure or distribution in the data in order to learn more about the data.

About this article

Publication history