Advertisement

Editorial Commentary: Personalized Hip Arthroscopy Outcome Prediction Using Machine Learning—The Future Is Here

      Abstract

      Machine learning and artificial intelligence are increasingly used in modern health care, including arthroscopic and related surgery. Multiple high-quality, Level I evidence, randomized, controlled investigations have recently shown the ability of hip arthroscopy to successfully treat femoroacetabular impingement syndrome and labral tears. Contemporary hip preservation practice strives to continually refine and improve the value of care provision. Multiple single-center and multicenter prospective registries continue to grow as part of both United States–based and international hip preservation–specific networks and collaborations. The ability to predict postoperative patient-reported outcomes preoperatively holds great promise with machine learning. Machine learning requires massive amounts of data, which can easily be generated from electronic medical records and both patient- and clinician-generated questionnaires. On top of text-based data, imaging (e.g., plain radiographs, computed tomography, and magnetic resonance imaging) can be rapidly interpreted and used in both clinical practice and research. Formidable computational power is also required, using different advanced statistical methods and algorithms to generate models with the ability to predict individual patient outcomes. Efficient integration of machine learning into hip arthroscopy practice can reduce physicians’ “busywork” of data collection and analysis. This can only improve the value of the patient experience, because surgeons have more time for shared decision making, with empathy, compassion, and humanity counterintuitively returning to medicine.
      We are living in an exciting and unprecedented time in the world of medicine—personalized, customized, patient-specific diagnoses and treatments are rendered every day. The approach to management of certain complex conditions (e.g., cancer, heart disease, diabetes, and arthritis) is now significantly influenced not only by their phenotype, what we see during routine clinical evaluation, but also by their genotype.
      • Vera A.M.
      • Peterson L.E.
      • Dong D.
      • et al.
      High prevalence of connective tissue gene variants in professional ballet.
      That’s right: People’s genes, their genetic makeup, their DNA, who they are—this information is being used not just for research but for clinical care.
      • Sochacki K.R.
      • Mather R.C.
      • Nwachukwu B.U.
      • et al.
      Sham surgery studies in orthopaedic surgery may just be a sham: A systematic review of randomized placebo-controlled trials.
      ,
      • Sochacki K.R.
      • Dong D.
      • Harris J.D.
      • et al.
      Author reply to "Placebo Trials in Orthopaedic Surgery" and "Review of Randomized Placebo-Controlled Trials.".
      Unfortunately, it is estimated that it takes 17 years for research evidence to reach clinical practice.
      • Morris Z.S.
      • Wooding S.
      • Grant J.
      The answer is 17 years, what is the question: Understanding time lags in translational research.
      However, the doubling time of medical knowledge was recently estimated at 73 days in 2020
      • Densen P.
      Challenges and opportunities facing medical education.
      ! Now, let’s closely examine the hip: If you combine the layered approach to the hip,
      • Draovitch P.
      • Edelstein J.
      • Kelly B.T.
      The layer concept: Utilization in determining the pain generators, pathology and how structure determines treatment.
      the Warwick agreement,
      • Griffin D.R.
      • Dickenson E.J.
      • O'Donnell J.
      • et al.
      The Warwick Agreement on femoroacetabular impingement syndrome (FAI syndrome): An international consensus statement.
      the Lisbon agreement,
      • Mascarenhas V.V.
      • Castro M.O.
      • Rego P.A.
      • et al.
      The Lisbon Agreement on femoroacetabular impingement imaging—Part 1: Overview.
      ,
      • Mascarenhas V.V.
      • Ayeni O.R.
      • Egund N.
      • et al.
      Imaging methodology for hip preservation: Techniques, parameters, and thresholds.
      and the Doha agreement,
      • Weir A.
      • Brukner P.
      • Delahunt E.
      • et al.
      Doha agreement meeting on terminology and definitions in groin pain in athletes.
      you will see that across the hip and pelvis, there are the following: 5 joints, 5 bones, 21 crossing musculotendinous units, the lumbosacral nerve plexus and its branches, the aortoiliac-femoral vascular system, dozens of different subjective symptoms and objective physical examination signs variably sensitive and specific for hip disease, 17 different plain radiographic views, 12 different magnetic resonance imaging or computed tomography series, 62 different imaging measurements, and 19 different hip-specific patient-reported outcome scores analyzed across 4 different broad measures of clinical relevance. Furthermore, if you acquire the average amount of data of a single patient entered in an electronic medical record, there are up to 32,000 discrete data points available for analysis.
      • Milinovich A.
      • Kattan M.W.
      Extracting and utilizing electronic health data from Epic for research.
      Thus, how can we possibly use all the available data from “the patient in front of you” to determine the best patient selection and surgical technique(s) to optimize outcomes and minimize complications and reoperations? “We” cannot do it—at least not alone. We need computers. We need processing power. We need the ability to manage massive amounts of data. Although hip preservation–specific databases are growing internationally (e.g., Academic Network of Conservational Hip Outcomes Research [ANCHOR], Multicenter Arthroscopy of the Hip Outcomes Research Network [MAHORN], Multicenter Arthroscopic Study of the Hip [MASH], Non-Arthroplasty Hip Registry [NAHR], Danish Hip Arthroscopy Registry [DHAR], North American Hip Arthroscopy Registry [NAHAR]), as are a number of insurance (e.g., PearlDiver and MarketScan) and geographic (e.g., New York statewide and National Surgical Quality Improvement Program [NSQIP]) databases, in “Development and Internal Validation of Supervised Machine Learning Algorithms for Predicting Clinically Significant Functional Improvement in a Mixed Population of Primary Hip Arthroscopy Patients,” Kunze, Polce, Nwachukwu, Chahla, and Nho
      • Kunze K.N.
      • Polce E.M.
      • Nwachukwu B.U.
      • Chahla J.
      • Nho S.J.
      Development and internal validation of supervised machine learning algorithms for predicting clinically significant functional improvement in a mixed population of primary hip arthroscopy patients.
      attempt to integrate machine learning and boldly go where no hip preservation study has gone before—customized point-of-care hip arthroscopy outcome prediction using a single surgeon’s personal database. Although Occam’s razor tells us that the simplest answer is usually the correct one, hip preservation surgery outcome prediction is far more complex than the law of parsimony.
      A simple PubMed search query of “machine learning orthopedic surgery” yields only 13 publications prior to 2015. Since then, the number of publications has exponentially increased: 16 in 2016, 24 in 2017, 43 in 2018, 108 in 2019, 187 in 2020, and an estimated 272 in 2021. Contemporary machine learning algorithms require large data sets, often sizes that a single surgeon cannot sufficiently populate to generate meaningful utility. The use of “big data” in orthopaedic surgery obviates these sample size issues—via either multicenter registries or insurance databases. Unfortunately, when aggregating large insurance-based databases, inclusion of low-volume surgeons’ patients may skew the data to significantly higher rates of complications and reoperations, not accurately reflecting optimal hip arthroscopy outcomes.
      • Sochacki K.R.
      • Jack II, R.A.
      • Safran M.R.
      • Nho S.J.
      • Harris J.D.
      There is a significant discrepancy between "big data" database and original research publications on hip arthroscopy outcomes: A systematic review.
      ,
      • Mehta N.
      • Chamberlin P.
      • Marx R.G.
      • et al.
      Defining the learning curve for hip arthroscopy: A threshold analysis of the volume-outcomes relationship.
      It has even been shown that the majority of hip arthroscopy procedures (85.1%) are performed by low-volume surgeons (those performing <30 such procedures per year).
      • Degen R.M.
      • Bernard J.A.
      • Pan T.J.
      • et al.
      Hip arthroscopy utilization and associated complications: A population-based analysis.
      Databases are only as good as the data entered into them. This is why the study by Kunze et al.
      • Kunze K.N.
      • Polce E.M.
      • Nwachukwu B.U.
      • Chahla J.
      • Nho S.J.
      Development and internal validation of supervised machine learning algorithms for predicting clinically significant functional improvement in a mixed population of primary hip arthroscopy patients.
      is so valuable. Different combinations and permutations of the senior author’s personal surgical database have been published nearly 2 hundred times. In his most recent iteration, patient-reported outcomes showed excellent mean scores (Hip Outcome Score–Activities of Daily Living [HOS-ADL], 95 ± 8; Hip Outcome Score–Sport-Specific Subscale, 89 ± 17; modified Harris Hip Score, 91 ± 11; and achievement of minimal clinically important difference [MCID] and patient acceptable symptom state in at least 1 outcome measure in 96% and 92% of patients, respectively) and the reoperation rate was 2.9% at 5 years’ follow-up.
      • Beck E.C.
      • Nwachukwu B.U.
      • Chahla J.
      • Clapp I.
      • Jan K.
      • Nho S.J.
      Complete capsule closure provides clinically significant outcome improvement and higher survivorship after hip arthroscopy at minimum 5-year follow-up.
      Similarly, among adolescents (aged 11-21 years), at 5 years’ follow-up, 88% of patients reached the MCID and there was a 2.4% reoperation rate.
      • Beck E.C.
      • Nwachuckwu B.U.
      • Jan K.
      • Nho S J.
      Hip arthroscopy for femoroacetabular impingement syndrome in adolescents provides clinically significant outcome benefit at minimum five year follow-up.
      Corroborating these outcomes with those of a United States–based statewide database, the lowest reoperation risk was achieved in surgeons with more than 518 career hip arthroscopy procedures (2.6% reoperation incidence at 5 years)
      • Mehta N.
      • Chamberlin P.
      • Marx R.G.
      • et al.
      Defining the learning curve for hip arthroscopy: A threshold analysis of the volume-outcomes relationship.
      or more than 164 (5% at 2 years, 10% at 5 years, and 10% at 10 years) to 340 (2% at 2 years, 2% at 5 years, and 2% at 10 years) annual hip arthroscopy procedures.
      • Degen R.M.
      • Pan T.J.
      • Chang B.
      • et al.
      Risk of failure of primary hip arthroscopy—A population-based study.
      Furthermore, the risk of complications and readmissions at 90 days was significantly lower for high-volume surgeons (readmission rates of 2.2% for <102 cases/year, 0.85% for 164-340 cases/year, and 0.58% for >340 cases/year).
      • Degen R.M.
      • Bernard J.A.
      • Pan T.J.
      • et al.
      Hip arthroscopy utilization and associated complications: A population-based analysis.
      Thus, clearly it is absolutely necessary to establish a “best-case scenario” when optimizing machine learning models by using data with the highest chance of success. This “ideal” situation approach establishes efficacy—can it work? The “real-world” situation soon follows and establishes effectiveness—will it work?
      This is not Kunze et al.’s first rodeo with hip preservation machine learning. They used TRIPOD (Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis) guidelines and 19 routinely collected preoperative features from 935 primary hip arthroscopy patients in a recent study (in which the primary outcome measure was meeting the MCID for the visual analog scale [VAS] score for satisfaction via an anchor-based method at 2 years’ follow-up).
      • Kunze K.N.
      • Polce E.M.
      • Rasio J.
      • Nho S.J.
      Machine learning algorithms predict clinically significant improvements in satisfaction after hip arthroscopy.
      By use of an 80:20 random sampling (common ratio for training to testing in machine learning), the training set (n = 749) underwent 10-fold cross validation 3 times to develop 5 supervised machine learning algorithms: random forest, stochastic gradient boosting, neural network, elastic-net penalized logistic regression, and support vector machine. The best algorithm’s performance was assessed via discrimination, calibration, the Brier score, and decision-curve analysis. The testing set (n = 186) was used for internal validation. The neural network was the best-performing algorithm. By use of this algorithm, the top 5 features predictive of not achieving the MCID for the VAS satisfaction score were the lateral center-edge angle, a history of anxiety or depression, 1 or more drug allergies, preoperative symptom duration of more than 2 years, and Workers’ Compensation status.
      In a separate investigation of the senior author’s surgical patients, Nwachukwu et al.
      • Nwachukwu B.U.
      • Beck E.C.
      • Lee E.K.
      • et al.
      Application of machine learning for predicting clinically meaningful outcome after arthroscopic femoroacetabular impingement surgery.
      generated predictive models for achieving the MCID in the modified Harris Hip Score, HOS-ADL, and Hip Outcome Score–Sport-Specific Subscale using feature selection with LASSO (least absolute shrinkage and selection operator) (which reduces a data set to its most meaningful features) and binary logistic regression with a generalized linear model. Predictors of not achieving the MCID in the analyzed patient-reported outcomes were anxiety or depression, preoperative symptoms for more than 2 years, higher body mass index (BMI), and preoperative hip injection. Clearly, a patient’s mental wellness plays a significant role in hip arthroscopy outcomes.
      • Sochacki K.R.
      • Brown L.
      • Cenkus K.
      • Di Stasi S.
      • Harris J.D.
      • Ellis T.J.
      Preoperative depression is negatively associated with function and predicts poorer outcomes after hip arthroscopy for femoroacetabular impingement.
      • Sochacki K.R.
      • Jack II, R.A.
      • Bekhradi A.
      • Delgado D.
      • McCulloch P.C.
      • Harris J.D.
      Are self-reported medication allergies associated with worse Hip Outcome Scores prior to hip arthroscopy?.
      • Lansdown D.A.
      • Ukwuani G.
      • Kuhns B.
      • Harris J.D.
      • Nho S.J.
      Self-reported mental disorders negatively influence surgical outcomes after arthroscopic treatment of femoroacetabular impingement.
      Kunze et al.
      • Kunze K.N.
      • Polce E.M.
      • Nwachukwu B.U.
      • Chahla J.
      • Nho S.J.
      Development and internal validation of supervised machine learning algorithms for predicting clinically significant functional improvement in a mixed population of primary hip arthroscopy patients.
      have now used their early work in hip arthroscopy and machine learning to take the colossal amount of available data and generate a “plug and play” equation for their patients—“a digital application using local explanations to provide customized risk assessment.” By use of 21 preoperative features from 818 patients who underwent primary hip arthroscopy (the primary outcome measure was meeting the MCID for the HOS-ADL via a distribution-based method at 2 years’ follow-up), an 80:20 training-testing ratio was used with 5 supervised machine learning algorithms with 3 iterations of 10-fold cross validation and the algorithms’ performance was assessed with 4 metrics. The stochastic gradient boosting model was the best algorithm, and there were 8 preoperative features deemed most important for predicting achievement of the MCID in the HOS-ADL: age, BMI, preoperative HOS-ADL, preoperative pain level, sex, Tönnis grade, symptom duration, and drug allergies. These features were used to create an algorithm for customized risk assessment (https://orthoapps.shinyapps.io/HPRG_ADL/).
      Now picture this: You see a new patient in your office, Jane Doe. She is a 23-year-old soccer player (BMI, 22 kg/m2) with a 3-year history of hip pain, an HOS-ADL score of 55 (of 100), and a VAS pain score of 70 (of 100), and she is allergic to penicillin, which causes a severe rash. Her plain radiographs reveal no arthritis (Tönnis grade 0). You enter this information onto the screen of the computer in the patient’s room as you go over her imaging. The screen then, in less than 1 second, generates a red-blue color-coded figure showing supporting or contradicting factors that determine the patient’s probability of achieving clinically significant improvement in function—85.9%. Her low HOS-ADL, high VAS pain score, female sex, and drug allergy portend a lower chance of success, whereas her normal BMI, young age, Tönnis grade, and symptom duration increase her chance of success. You then click the comparison tab on the screen and compare her probability of success with that of an “average” patient undergoing hip arthroscopy. This allows you to have a truly “informed” informed-consent, shared decision-making preoperative discussion. The efficiency gained in this real-world scenario permits you more time in the office to discuss the patient’s postoperative goals and expectations, focus more on education and understanding, answer questions and concerns, and include her care team (e.g., family, friends, physical therapist, employer, and coworkers) in the discussion to optimize the value of her care experience. She then leaves the office and tells her parents and her spouse about how great her surgeon was, using artificial intelligence to predict her outcome, and she goes to Google and leaves you a 5-star review with a glowing endorsement of the quality of the visit.
      Sounds great, right? A few caveats must be mentioned in embracing this new aspect of your everyday clinical practice. First and foremost, the chance of success is based on the senior author’s patients, not your own. His practice is high volume (career and annual maintenance), with proof of success in the published literature. Thus, extrapolation to your practice requires careful interpretation. The disclaimer section of the application’s webpage clearly states that it is for “general educational purposes only” and “is intended to serve as a supplement to, and not a substitute for the knowledge, expertise, skill and judgment of health care professionals.” The latter is such an incredibly important point—It is a supplement, not a replacement. Computers will never replace doctors; they will help us. Despite the “black box phenomenon”—the inner nodes (neurons) running the algorithm working between data input and output—the humanity, empathy, and compassion cannot be replicated. Eric Topol, M.D., cardiologist, geneticist, number 1 best-selling book author, and digital medicine researcher, has written 3 wonderful, highly relevant texts that I fully recommend as a complement to this editorial commentary and the study by Kunze et al.
      • Kunze K.N.
      • Polce E.M.
      • Nwachukwu B.U.
      • Chahla J.
      • Nho S.J.
      Development and internal validation of supervised machine learning algorithms for predicting clinically significant functional improvement in a mixed population of primary hip arthroscopy patients.
      : “Deep Medicine: How Artificial Intelligence Can Make Healthcare Human Again,”
      • Topol E.
      Deep medicine: How artificial intelligence can make healthcare human again.
      “The Patient Will See You Now: The Future of Medicine Is in Your Hands,”
      • Topol E.
      The patient will see you now: The future of medicine is in your hands.
      and “The Creative Destruction of Medicine: How the Digital Revolution Will Create Better Health Care.”
      • Topol E.
      The creative destruction of medicine: How the digital revolution will create better health care.
      As both Topol and our fictional patient, Jane Doe, have shown, “the transformational potential of AI [artificial intelligence] is in its power to enhance the human aspect of medicine, which is something we’ve lost.”
      A doctor explains how artificial intelligence could improve the patient-doctor bond. Eric Topol’s Deep Medicine paints an optimistic view of the future of algorithms and medicine.
      Machine learning as well as artificial intelligence “creates space for the real healing that occurs between a doctor who listens and a patient who needs to be heard. The counterintuitive recognition that technology can create space for compassion in the clinical setting could mean fewer burned-out doctors, more empowered patients, cost savings, and an entirely new way to approach medicine.”
      Deep medicine
      How artificial intelligence can make health care human again.
      Another caveat is that this technology is currently not free; time, effort, money, and patients are required to generate the database. The senior author’s 3 articles hitherto mentioned all used a cloud-based platform (Patient IQ
      Patient IQ.
      ) with built-in machine learning infrastructure using analytical predictive statistics—a truly “data-driven medicine.”
      • Kunze K.N.
      • Polce E.M.
      • Nwachukwu B.U.
      • Chahla J.
      • Nho S.J.
      Development and internal validation of supervised machine learning algorithms for predicting clinically significant functional improvement in a mixed population of primary hip arthroscopy patients.
      ,
      • Kunze K.N.
      • Polce E.M.
      • Rasio J.
      • Nho S.J.
      Machine learning algorithms predict clinically significant improvements in satisfaction after hip arthroscopy.
      ,
      • Nwachukwu B.U.
      • Beck E.C.
      • Lee E.K.
      • et al.
      Application of machine learning for predicting clinically meaningful outcome after arthroscopic femoroacetabular impingement surgery.
      Although not all surgeons who perform arthroscopic and related surgery need to be machine learning experts, the future is here now. You don’t need to know why principal component analysis is different from shrinkage and LASSO (least absolute shrinkage and selection operator) or why a deep neural network is different from a stochastic gradient boosting model, but you should have a basic understanding of the potential real utility for your day-to-day clinical practice. A complete discussion of the simplest basics to the most complex details is far beyond the scope of this editorial commentary. However, for interested readers, please consult PubMed for a few experts in orthopaedic surgery machine learning and artificial intelligence: Prem Ramkumar, Ben Nwachukwu, and Kyle Kunze. Arthroscopy recently published a terrific narrative review discussing the definitions, applications, and limitations of artificial intelligence and machine learning.
      • Ramkumar P.N.
      • Kunze K.N.
      • Haeberle H.S.
      • et al.
      Clinical and research medical applications of artificial intelligence.

      Supplementary Data

      References

        • Vera A.M.
        • Peterson L.E.
        • Dong D.
        • et al.
        High prevalence of connective tissue gene variants in professional ballet.
        Am J Sports Med. 2020; 48: 222-228
        • Sochacki K.R.
        • Mather R.C.
        • Nwachukwu B.U.
        • et al.
        Sham surgery studies in orthopaedic surgery may just be a sham: A systematic review of randomized placebo-controlled trials.
        Arthroscopy. 2020; 36: 2750-2762.e2
        • Sochacki K.R.
        • Dong D.
        • Harris J.D.
        • et al.
        Author reply to "Placebo Trials in Orthopaedic Surgery" and "Review of Randomized Placebo-Controlled Trials.".
        Arthroscopy. 2020; 36: 2779-2784
        • Morris Z.S.
        • Wooding S.
        • Grant J.
        The answer is 17 years, what is the question: Understanding time lags in translational research.
        J R Soc Med. 2011; 104: 510-520
        • Densen P.
        Challenges and opportunities facing medical education.
        Trans Am Clin Climatol Assoc. 2011; 122: 48-58
        • Draovitch P.
        • Edelstein J.
        • Kelly B.T.
        The layer concept: Utilization in determining the pain generators, pathology and how structure determines treatment.
        Curr Rev Musculoskelet Med. 2012; 5: 1-8
        • Griffin D.R.
        • Dickenson E.J.
        • O'Donnell J.
        • et al.
        The Warwick Agreement on femoroacetabular impingement syndrome (FAI syndrome): An international consensus statement.
        Br J Sports Med. 2016; 50: 1169-1176
        • Mascarenhas V.V.
        • Castro M.O.
        • Rego P.A.
        • et al.
        The Lisbon Agreement on femoroacetabular impingement imaging—Part 1: Overview.
        Eur Radiol. 2020; 30: 5281-5297
        • Mascarenhas V.V.
        • Ayeni O.R.
        • Egund N.
        • et al.
        Imaging methodology for hip preservation: Techniques, parameters, and thresholds.
        Semin Musculoskelet Radiol. 2019; 23: 197-226
        • Weir A.
        • Brukner P.
        • Delahunt E.
        • et al.
        Doha agreement meeting on terminology and definitions in groin pain in athletes.
        Br J Sports Med. 2015; 49: 768-774
        • Milinovich A.
        • Kattan M.W.
        Extracting and utilizing electronic health data from Epic for research.
        Ann Transl Med. 2018; 6: 42
        • Kunze K.N.
        • Polce E.M.
        • Nwachukwu B.U.
        • Chahla J.
        • Nho S.J.
        Development and internal validation of supervised machine learning algorithms for predicting clinically significant functional improvement in a mixed population of primary hip arthroscopy patients.
        Arthroscopy. 2021; 37: 1488-1497
        • Sochacki K.R.
        • Jack II, R.A.
        • Safran M.R.
        • Nho S.J.
        • Harris J.D.
        There is a significant discrepancy between "big data" database and original research publications on hip arthroscopy outcomes: A systematic review.
        Arthroscopy. 2018; 34: 1998-2004
        • Mehta N.
        • Chamberlin P.
        • Marx R.G.
        • et al.
        Defining the learning curve for hip arthroscopy: A threshold analysis of the volume-outcomes relationship.
        Am J Sports Med. 2018; (363546517749219)
        • Degen R.M.
        • Bernard J.A.
        • Pan T.J.
        • et al.
        Hip arthroscopy utilization and associated complications: A population-based analysis.
        J Hip Preserv Surg. 2017; 4: 240-249
        • Beck E.C.
        • Nwachukwu B.U.
        • Chahla J.
        • Clapp I.
        • Jan K.
        • Nho S.J.
        Complete capsule closure provides clinically significant outcome improvement and higher survivorship after hip arthroscopy at minimum 5-year follow-up.
        Arthroscopy. 2021; ([published online January 30, 2021])
        • Beck E.C.
        • Nwachuckwu B.U.
        • Jan K.
        • Nho S J.
        Hip arthroscopy for femoroacetabular impingement syndrome in adolescents provides clinically significant outcome benefit at minimum five year follow-up.
        Arthroscopy. 2021; 37: 1467-1473
        • Degen R.M.
        • Pan T.J.
        • Chang B.
        • et al.
        Risk of failure of primary hip arthroscopy—A population-based study.
        J Hip Preserv Surg. 2017; 4: 214-223
        • Kunze K.N.
        • Polce E.M.
        • Rasio J.
        • Nho S.J.
        Machine learning algorithms predict clinically significant improvements in satisfaction after hip arthroscopy.
        Arthroscopy. 2021; 37: 1143-1151
        • Nwachukwu B.U.
        • Beck E.C.
        • Lee E.K.
        • et al.
        Application of machine learning for predicting clinically meaningful outcome after arthroscopic femoroacetabular impingement surgery.
        Am J Sports Med. 2020; 48: 415-423
        • Sochacki K.R.
        • Brown L.
        • Cenkus K.
        • Di Stasi S.
        • Harris J.D.
        • Ellis T.J.
        Preoperative depression is negatively associated with function and predicts poorer outcomes after hip arthroscopy for femoroacetabular impingement.
        Arthroscopy. 2018; 34: 2368-2374
        • Sochacki K.R.
        • Jack II, R.A.
        • Bekhradi A.
        • Delgado D.
        • McCulloch P.C.
        • Harris J.D.
        Are self-reported medication allergies associated with worse Hip Outcome Scores prior to hip arthroscopy?.
        Arthroscopy. 2018; 34: 1856-1861
        • Lansdown D.A.
        • Ukwuani G.
        • Kuhns B.
        • Harris J.D.
        • Nho S.J.
        Self-reported mental disorders negatively influence surgical outcomes after arthroscopic treatment of femoroacetabular impingement.
        Orthop J Sports Med. 2018; 6 (2325967118773312)
        • Topol E.
        Deep medicine: How artificial intelligence can make healthcare human again.
        Basic Books, Hachette Book Group, New York2019
        • Topol E.
        The patient will see you now: The future of medicine is in your hands.
        Basic Books, Hachette Book Group, New York2016
        • Topol E.
        The creative destruction of medicine: How the digital revolution will create better health care.
        Basic Books, Perseus Books Group, New York2013
      1. A doctor explains how artificial intelligence could improve the patient-doctor bond. Eric Topol’s Deep Medicine paints an optimistic view of the future of algorithms and medicine.
        (Published 2019)
        • Deep medicine
        How artificial intelligence can make health care human again.
        (Published 2019)
      2. Patient IQ.
        (Published 2021)
        https://www.patientiq.io/
        Date accessed: February 14, 2021
        • Ramkumar P.N.
        • Kunze K.N.
        • Haeberle H.S.
        • et al.
        Clinical and research medical applications of artificial intelligence.
        Arthroscopy. 2021; 37: 1694-1697