当前位置： >>
DATA MINING FOR HEALTHCARE
MANAGEMENT 2011
DATA MINING FOR HEALTHCARE MANAGEMENTPrasanna Desikan
Center for Healthcare Innovation Allina Hospitals and Clinics USA Kuo-Wei Hsu Jaideep Srivastava kuowei.hsu@gmai
< srivasta@cs.umn.edu National Chengchi University University of Minnesota & Taiwan Center for Healthcare Innovation Allina Hospitals and Clinics USAOutline? Introduction ? Why Data Mining can aid Healthcare ? Healthcare Management Directions ? Overview of Research ? Kinds of Data ? Challenges in data mining for healthcare ? Framework ? Prominent Models ? Sample case study ? Summary and Future Directions4/29/20112INTRODUCTIONHealthcare Management“Health administration or healthcare administration is the field relating to leadership, management, and administration of hospitals, hospital networks, and health care systems.”* It is actually a broad area that could encompass:-Healthcare Informatics -Medical Device Industry -Pharmaceutical Industry -Hospital Management -System Biologyand many more….*http://en.wikipedia.org/wiki/Healthcare_management 4/29/2011 4Healthcare Ecosystem C A PerspectiveGovt. Healthcare Department Health InsuranceCenters for Disease ControlClinicPatient PharmacyPharmacy Benefit ManagementDialysis Dental ChiropracticSpecialized ServicesMedical EquipmentHospitalDrug Industry4/29/20115Interface between Patients and Medical ServicesReferral4/29/20116Motivation for Healthcare Management? Problem:C “Government health spending wastes a heck of a lot of money,” U.S. Vice President Joe Biden, 2/25/2010 C “Healthcare spending 17 percent of economy”, , 2/4/2010 C “More than $1.2 trillion spent on health care each year is a waste of money”, , 8/10/20094/29/20117Performance of the U.S. Health Care System InternationallyReflects lack of leverage of Information technology? ??U.S. is lagging in adoption of national policies that promote primary care, quality improvement, and information technology. Health reform legislation addresse for instance, the American Recovery and Reinvestment Act signed by President Obama in February 2009 included approximately $19 billion to expand the use of health information technology. The Patient Protection and Affordable Care Act of 2010 also will work toward realigning providers&#39; financial incentives, encouraging more efficient organization and delivery of health care, and investing in preventive and population health.Key Takeaway: Need for an efficient , organized and knowledge based decision support systems.Source : monwealthfund.org/Content/Publications/Fund-Reports/2010/Jun/Mirror-Mirror-Update.aspx.4/29/20118WHY DATA MINING CAN AID HEALTHCAREWhy Data Mining?? Healthcare industry today generates large amounts ofcomplex data about patients, hospitals resources, disease diagnosis, electronic patient records, medical devices etc. ? The large amounts of data is a key resource to be processed and analyzed for knowledge extraction that enables support for cost-savings and decision making. ? Data mining? brings a set of tools and techniques that can be applied to thisprocessed data to discover hidden patterns ? that provide healthcare professionals an additional source of knowledge for making decisions? The decisions rests with health care professionals.4/29/2011 10How does data mining help?? Data mining is a collection of algorithmic ways to extract informativepatterns from raw dataC Data mining is purely data- this feature is important in health care? y = f(x)? We have seen x (set of independent variables) and observed y (dependentvariable); data mining tells us something about the nature of f ? x = symptoms or test results, y = ? x = treatments, y = symptom ? It tells us “how”? How is x related to y? What function describes their relationship?? f(x, y) = score, or f(x|y) = Pr(x|y) ? Data mining does not (directly) explain to us “why” C Why does xcause y?? It helps doctors/physicians (domain experts) figure that (causation) out ? ‘Descriptive/predictive model’ vs. ‘Causal model’4/29/2011 11Who does it benefit?? Data mining can help? Healthcare insurers detect fraud and abuse, ? Healthcare organizations make customerrelationship management decisions, ? Physicians identify effective treatments and best practices, and ? Patients receive better and more affordable healthcare services.4/29/201112HEALTHCARE MANAGEMENT DIRECTIONSKey Dimensions in Healthcare Management [Koh05]? Diagnosis and Treatment ? Healthcare Resource Management ? Customer Relationship Management ? Fraud and Anomaly Detection4/29/201114Diagnosis and TreatmentMedical decision support (to doctors) [Hardin2008]?Analysis of digitized images of skin lesions to diagnosemelanoma [Burroni 2004] ?Computer-assisted texture analysis of ultrasound images aids monitoring of tumor response to chemotherapy[Hub2000] ?Predicting the presence of brain neoplasm with magnetic resonance spectroscopy [Zellner2004] ?Analysis of digital images of tissue sections to identify and quantify senile plagues for diagnosing and evaluating the severity of Alzheimer’s disease. [Hibbard1997]4/29/201115Diagnosis and TreatmentTreatment plan (to patients)?Data mining could be particularly useful in medicine whenthere is no dispositive evidence favoring a particular treatment option ?Based on patients’ profile, history, physical examination, diagnosis and utilizing previous treatment patterns, new treatment plans can be effectively suggested ?Examples? Onset, treatment and management of depression [Hadzic2010] ? Treatment Decision Support Tool for Patients with Uterine Fibroids[Campbell2010]4/29/201116Healthcare Resource Management? Using logistic regression models to comparehospital profiles based on risk-adjusted death with 30 days of non-cardiac surgery ? Neural network system to predict the disposition in children presenting to the emergency room with bronchiolitis ? Predicting the risk of in-hospital mortality in cancer patients with nonterminal disease4/29/201117Healthcare Resource ManagementPrediction of inpatient length of stay? Effectively manage the resource allocation by identifyinghigh risk areas and predicting the need and usage of various resources. ? For example, a key problem in the healthcare area is the measurement of flow of patients through hospitals and other health care facilities. ? If the inpatient length of stay (LOS) can be predicted efficiently, the planning and management of hospital resources can be greatly enhanced.4/29/201118Customer Relationship Management? CRM is to establish close customer relationships[Rygielski02]? The focus shifts away from the breadth of customer base (product-oriented view, mass marketing) to the depth of each customer’s needs (customer-oriented view, one-to-one marketing)? CRM is built on an integrated view of the customer acrossthe whole organization [Puschmann01]? Customers have a fractured v the enterprisehas a splintered view of the customer? Kohli et al. demonstrate a web-based Physician ProfilingSystem (PPS) to strengthen relationships with physicians and improve hospital profitability and quality [Kohli01]4/29/2011 19Customer Relationship Management (cont.)? Development of total customer relationship in healthcare includesseveral tenets [Berwick97]? “In a helping profession, the ultimate judge of performance is the person ? ? ? ?helped” “Most people, including sick people, are reasonable most of the time” “Different people have different, legitimate needs” “Pain and fear produce anxiety in both the victim and the helper” “Meeting needs without waste is a strategic and moral imperative”? Some demographic characteristics (e.g. age, health status, and race)and institutional characteristics (e.g. hospital size) consistently have a significant effect on a patient’s satisfaction scores [Young00] ? Chronic illnesses (e.g. diabetes and asthma) require selfmanagement and a collaborative patient-physician relationship [Ouschan06]4/29/2011 20Customer Relationship Management (cont.)? The principles of applying of data mining for customerrelationship management in the other industries are also applicable to the healthcare industry. ? The identification of usage and purchase patterns and the eventual satisfaction can be used to improve overall customer satisfaction. ? The customers could be patients, pharmacists, physicians or clinics. ? In many cases prediction of purchasing and usage behavior can help to provide proactive initiatives to reduce the overall cost and increase customer satisfaction.4/29/2011 21Fraud and Anomaly Detection? Bolton and Hand briefly discuss healthcare insurancefraud [Bolton02]? Examples of frauds:? Prescription fraud: claims for patients who do not exist ? Upcoding: claims for a medical procedure which is moreexpensive or not performed at all ? Examples of detection methods: ? Neural networks, genetic algorithms, nearest neighbor methods ? Comparing observations with those they have similar geodemographics4/29/201122Fraud and Anomaly Detection? Data mining has been used very successfully in aiding theprevention and early detection of medical insurance fraud. ? The ability to detect anomalous behavior based on purchase, usage and other transactional behavior information has made data mining a key tool in variety of organizations to detect fraudulent claims, inappropriate prescriptions and other abnormal behavioral patterns. ? Another key area where data mining based fraud detection is useful is detection and prediction of faults in medical devices.4/29/201123Examples of Research in Data Mining for Healthcare Management.Researching topic Researching institute DatasetGeriatric Medicine department of a metropolitan teaching hospital in the UK. Healthcare data mining: School of Information predicting inpatient length of Management and Engineering, stay Shanghai U Harrow School of Computer Science The Center for Computational Designing Patient-Specific Seizure Detectors Learning Systems (CCLS) and The From Multiple Frequency Columbia Bands of Intra-cranial University Medical School (CUMC) EEG Using Support Vector Machines Classification, Treatment and MGR University, C UVCE , Management of Alzheimer’s Bangalore,; Defence Institue of Disease Using Various Advanced Technology Pune Machine Learning MethodsColumbia University Medical School has collected approximately 30 TB of intracranial EEG recordings. National Institute on Aging, USA.4/29/201124OVERVIEW OF RESEARCHKinds of Data? “An EHR is an electronic version of a patient’s medicalhistory, that is maintained by the health- care provider over time, and includes all of the key administrative clinical data relevant to that person’s care under a particular provider, including demographics, progress notes, problems, medications, vital signs, past medical history, immunizations, laboratory data, medical images and radiology reports. “ [Maglogiannis09] ? HL7 (Health Level Seven)? Developed to improve health informatics interoperability ? Working in the 7th layer, application layer, of the Open SystemsInterconnection model ? www.hl7.org4/29/2011 26EMR and EHR? Electronic Medical Record (EMR): It contains patientinformation that is stored and retrieved locally in a standalone system used by a provider ? Electronic Health Record (EHR): It contains patient information that is stored and retrieved in systems used by all providers who care about the patient ? Many use EMR and EHR interchangeably ? Examples of applying data mining on EMR/EHR:? Ludwick and Doucette study the adaption of EMR in primary care[Ludwick09] ? Cerrito works on EMR from an Emergency Department [Cerrito07] ? Buczak et al. works on disease surveillance on EMR [Buczak09]4/29/2011 27Health Level 7 (HL 7)? HL7 provides standards for interoperability that improve care delivery,optimize workflow, reduce ambiguity and enhance knowledge transfer among healthcare providers [HL7]The V2.4 MessageThe V3 Message4/29/2011284/29/201129Kinds of Data[Maglogiannis09]Kinds of Data? Electronic Nursing Record (ENR): “While improving healthcare practices and patient care, it also provides easily and rapidly available data for a decision support system in real time.” [Santos09]“ … the medical information sources being distributed, heterogeneous and complex …” [Santos09]4/29/201130Kinds of Data? Data warehouse for integration of “evidence-based” datasources [Stolba06][Stolba06]4/29/2011[Stolba06]31Challenges in Data Mining for Healthcare? Data sets from various data sources [Stolba06] ? Example 1: Patient referral data can vary extensivelybetween cases because structure of patient referrals is up to general practitioner who refers the patient [Persson09] ? Example 2: Catley et al. use neural networks to predict preterm birth on a heterogeneous maternal population [Catley06] ? Example 3: “Traditional clinical-based prognosis models were discovered to contain some restrictions to address the heterogeneity of breast cancer” [Ahmad09]4/29/2011 32Challenges in Data Mining for Healthcare? Data from heterogeneous sources presentchallenges [Kwiatkowska07]? Sampling bias: “Clinical studies use diverse collecting ? ? ? ?methods, inclusion criteria, and sampling methods” Referral bias: “Data represent a preselected group with a high prevalence of disease” Selection bias: “Clinical data sets include patients with different demographics” Method bias: “Predictors have varied specifications, granularities, and precisions” Clinical spectrum bias: “Patient records represent varied severity of a disease and co-occurrence of other medical problems”334/29/2011Challenges in Data Mining for Healthcare? Missing values, noise, and outliers ? “Cleaning data from noise and outliers and handlingmissing values, and then finding the right subset of data, prepares them for successful data mining” [Razavi07] ? Transcription and manipulation of patient records often result in a high volume of noise and a high portion of missing values [O’Sullivan08] ? “Missing attribute values can impact the assessment of whether a particular combination of attribute-value pairs is significant within a dataset” [Laxminarayan06]4/29/201134Typical Data Mining Framework[Stolba06]4/29/201135Prominent Models? O’Sullivan et al. propose to incorporate formalizedexternal expert knowledge in building a prediction model for asthma exacerbation severity for pediatric patients in the emergency department [O’Sullivan08] ? The secondary knowledge source identified as relevant for our retrospective asthma data is the Preschool Respiratory Assessment Measure (PRAM) asthma index[O’Sullivan08](Children’s Hospital of Eastern Ontario, Ottawa, Canada)[O’Sullivan08]4/29/201136Prominent Models? Palaniappan and Awang demonstrate a web-basedIntelligent Heart Disease Prediction System (IHDPS) to use medical profiles to predict the likelihood of patients getting a heart disease [Palaniappan08]? Techniques: Decision trees, na?ve Bayes, neural networks? Each has its unique strength[Palaniappan08][Palaniappan08]4/29/201137Prominent Models? Persson and Lavesson investigate prediction models forpatient referrals [Persson09]? “A patient referral contains information that indicates the need forhospital care and this information is differently structured for different medical needs”4/29/201138Prominent Models? Kuttikrishnan et al. propose a system to assist cliniciansat the point of care [Kuttikrishnan10]? Knowledge base: Rules and associations of compiled data ? Inference engine: Combination of rules and patient’s data ? Mechanism to communicate: System-user interaction ? Technique: Neural networks[Kuttikrishnan10]4/29/2011[Kuttikrishnan10]39Prominent Models? De Toledo et al. discuss models of outcome prediction forsubarachnoid hemorrhage (SAH) [DeToledo09]? Techniques: C4.5, fast decision tree learner, partial decision trees,repeated incremental pruning to produce error reduction, nearest neighbor with generalization, and ripple down rule learner? The best classifier is the C4.5 algorithm[DeToledo09]4/29/201140Prominent Models? Razavi et al. discuss a model to predict recurrence ofbreast cancer [Razavi07]? “Identifying high-risk patients is vital in order to provide them withspecialized treatment” ? Technique: Decision tree[Razavi07]4/29/201141Prominent Models? Catley et al. propose a screen tool to early prediction ofpreterm birth [Catley06]? Current procedure uses costly and invasive clinical testing8 obstetrical variables were selected as being nonconfounding for predicting PTB: ? maternal age, ? number of babies this pregnancy, ? number of previous term babies, ? number of previous preterm babies, ? parity (total number of previous children), ? baby’s gender, ? whether mother has intention to breastfeed, maternal smoking after 20 weeks gestation. Some results: ? Previous term birth is not a good indicator of future preterm birth. ? Maternal intention to breastfeed has minimal impact on results.4/29/2011[Catley06][Catley06]42Prominent Models? Ahmad et al. propose to integrate clinical and microarraydata for accurate breast cancer prognosis [Ahmad09]? “Breast cancer patients with the same diagnostic and clinicalprognostics profile can have markedly different clinical outcomes”4/29/2011[Ahmad09]43Prominent Models? Chattopadhyay et al. discuss data-processing techniquesfor suicidal risk evaluation [Chattopadhyay08]? “Noise is eliminated incorporating the expertise of psychiatrists andpsychologists case-by-case” ? “Missing values are filled up with the most common value of the corresponding attributes” ? “Correlation analysis has been done to identify which two data within an attribute are statistically similar”[Chattopadhyay08]4/29/201144Prominent Models? Laxminarayan et al. propose a modified association rulemining technique to extract patterns from sequencevalued attributes such as sleep-related data by aggregating set of events that occur in the window [Laxminarayan06]? “Association mining may also be useful for selection of variablesprior to using logistic regression”[Laxminarayan06]4/29/201145Prominent Models? McGarry et al. investigate graph-mining techniques for protein-proteininteraction for diabetes research [McGarry07]? “The difficulties of complexity encountered by the pharmaceutical industrywhen developing the necessary assays for drug discovery have proved this conclusively, that there is no simple or direct link from genome to drugscular relationships.” [McGarry07]4/29/2011[McGarry07]46SAMPLE CASE STUDYCase Study1: Data Mining Based Decision Tool for Evaluating Treatment Choices for Uterine Fibroids [Campbell2010]Objectives? Can data mining techniques be applied to data collected frompatients with uterine fibroids in order to predict a treatment choice? ? Which data mining method is most successful in predicting the treatment decisions?Background C Uterine Fibroids?Non cancerous tumor in muscular layer of the uterus ?30-40% of women diagnosed ?Symptoms (vary significantly) C 50% of fibro Heavy and A Painfu Urinary f Infertility ?Size and symptoms may subside after menopause4/29/2011 48Treatment OptionsProcedure Hysterectomy Myomectomy Uterine Artery Emolization Hormone Therapy Watchful WaitingDescription Surgical removal of the uterus involves hospital stay and lengthy recovery period. Removal of one or more of the fibroids with open ab. Surgery or laparoscopic or endoscopic techniques. The uterine artery is blocked
the fibroid is starved of its blood supply. A drug treatment that No treatment. On going causes fibroid shrinkage. monitoring.AdvantagesSymptom relief with Non-surgical, Permanent solution because the Preserves the uterus and shorter hospital stay than conservative method of uterus is removed. cervix. hysterectomy or fibroid treatment. myomectomy.Sometimes fibroid symptoms diminish with menopause.DisadvantagesRisks include radiation, Temporary. Causes menopause, serious Reproductive potential is lost. Reoccurrence of fibroid menopausal symptoms. Fibroids may continue to infection, bleeding, Other side effects possible. symptoms possible if new May result in rapid return grow with an increase in embolization of other Recovery time of several weeks. fibroids grow. of symptoms after symptoms. organs, hysterectomy, treatment. loss ovarian function.Return to Normal Activity Hospital Time Procedure Time28-56 Days 2-5 Days 1.5-3 Hours4-44 Days 1-3 Days 1-3 Hours10 Days 1 Day .75-2 Hours____ 0 0____ 0 04/29/201149Treatment Decision? Patient ultimately makes the decision ? Many factors contribute to decision process C Pain, age, discomfort, sexual side effects, desire for children, etc. ? Research has not revealed any particular treatment as a“best practice”? There may not be a consistent way to measure “success”? Human analysis of data for decision making can beflawedC Personality, anecdotal information? Data mining could be used to direct women towardssuccessful treatment optionsC Which treatments have women like you been happy with in the past?Which have they been unhappy with?4/29/2011 50Data? Given that we do not have long term satisfaction scores,the scope of the project has been limited to the question:C “Which treatment have women like you chosen to pursue in thepast?”? Data C 171 Patients C 70 attributes (many redundant or metadata) C Data from a survey taken by patients identified as having recently made a decision about uterine fibroids C 8 different clinics in Minnesota between Feb 2007 and Feb 20084/29/201151Data4/29/201152Data4/29/201153Treatment Choices4/29/201154One Vs. Rest Scheme? Improved accuracy ? Rare class problem ? Global accuracy does not indicate performance from minorityclass perspective? Example: C In predicting hormone treatment:A 51 7 B 0 0 Classification A = Not Hormone Therapy B = Hormone TherapyC 87.931 accuracy C Recall (minority class perspective) = 04/29/201155Top 15 attributes as ranked by infogain for each treatmentRank 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 Hormone Therapy Attr. Val PAM &6 PAM &1 PPT &10 QH * PAM &7 PAM &4 PAM &5 PAM &2 QT * PAM &3 PAM &8 BB 8 QN 4 PAM =10 AGE 35 Hysterectomy Myomectomy Attr. Value PPT &9 PPT =10 PPT &7 PPT &8 PPT &3 PPT &5 IFD &4 IFD &6 BB 8 MB 7 PRR &8 IFD &2 PPT &6 IFD &1 PRR &6 Attr. PHB PHB PHB PHB PHB BUF PHB Race PHB BB PHB DUR PHB AGE PHB No Treatment Value Attr. Value &7 PRR &6 &8 PRR &7 &9 PRR &8 =10 PRR &5 &6 IFD &6 8 PRR &9 &4 IFD &2 * PRR &1 &5 IFD =0 1 IFD &1 &3 PRR &5 2 MB &4 &1 IFD &3 &50 MB &5 &2 BB &1 UAE Attr. BI BI DUR BI PSR IFD BB BI BI BI PSR BB DUR BI AGE Va &2 &1 3 &3 7 5 4 &1 &8 &9 6 3 1 &7 50 PAM PPT QH QT BB QN IFD PHB MB PRR BUF DUR BI PSR AGE Key Attribute Prefer to avoid medication Prefer permanent treatment Answer to hysterectomy quiz question Answer to treatment quiz question Bothered by bleeding Number quiz questions correct (5 total) Interferes with daily activities Prefer to have a baby Score of most bothering symptom Prefers rapid relief Bothered by urinary frequency Duration Bothered by infertility Prefer a short recuperation Age of the patient=, &, or & relationships are inferred with expert knowledge but not indicated with an infogain feature Example: A patient who scored their preference to have a baby as 7 or 56 4/29/2011 greater will be more likely to choose myomectomy for a treatmentTop 3 performing algorithm-ensemble-attribute combinations by 3 different metricsTreatments UAE Performance M etric Rank 1 2 F-M easure 3 1 2 Accuracy (%) 3 1 2 Area Under ROC 3 4.10% Algorithm A NB (30) A NB (35) S NB (30) A NB (35) A NB (30) A J48 (35) A NB (30) A NB (25) B NB (35) M etric 0.3 0. 94.86 94.42 0.3 0.9651 M yomectomy 11.10% Algorithm A NB (30) A NB (25) A NB (25) A J48 (50) A SC (50) A JR (100) B NB (30) B NB (35) S NB (30) M etric 0.9 0. 87.61 87.25 0.2 0.8375 Hormone 11.70% Algorithm A NB (25) S NB (50) B NB (50) A M P (35) A SC (30) A SC (25) S NB (25) S NB (30) B NB (25) M etric 0.6 0. 87.14 87.14 0.3 0.855 Hysterectomy 29.20% Algorithm B J48 (15) S J48 (15) B SC (15) B JR (35) A JR (35) B M P (15) S NB (35) B NB (25) S NB (15) M etric 0.691 0.2 79.63 78.99 78.7 0.87 0.7 No Treatment 43.90% Algorithm B J48 (15) B J48 (20) S J48 (15) B J48 (15) B J48 (20) S J48 (15) S NB (15) B NB (15) B NB (20) M etric 0.2 0. 78.55 78.12 0.852 0.Abbreviation = Ensemble + Algorithm + (#attributes used)? ? ?Ensemble Techniques A = Adaboost B = Bagging S = Simple (no ensemble)4/29/2011? ? ? ?Algorithms NB = Na?ve Bayes SC = Simple Cart MP = Multiperceptron JR = Jrip57Observations and Conclusions Observations? Worse performance with too few or too many attributes ? Peak is at greater attribute number for rare classes ? Rare classes or Na?ve Bayes more sensitive toConclusions?Classifiers are makingnumber of attributes0.9 0.8 0.7No Treatment (B J48)0.6F-Measure0.5 0.4 0.3Hysterectomy (B J48)Hormone (A NB)Myomectomy (A NB)0.2 0.1 0 10 15 20 25 30 35 50 100 208 Top k ranked attributes by infogain 4/29/2011UAE (A NB)predictions that are better than random and would improve with more data ?Which classification algorithms are most successful in predicting the treatment decisions for the patients? ?Rare Classes: ? Adaboost-Na?ve Bayes (2530 attributes ) selected ?Common Classes ? Bagging-J48 (15 attributes)58SUMMARY AND FUTURE DIRECTIONSSummary? An introduction to healthcare management and themotivation to study this field and its impact on current research and market trends. ? Research discussion? types of available data, ? the challenges involved ? prominent models.? A sample case study is presented to demonstrate how acertain application challenge can be addressed and the value of using data mining as a tool.4/29/201160Future Directions? Data Storage and Access ? Data collection and analysis ? Integration of models from other domains ? Theoretical and Applied Research.4/29/201161Data Storage and Access? Ensuring standard formats evolve and areadapted readily. ? Knowledge extraction by integrating data from various sources and formats. ? Storage and Access Mechanisms? Handling Dynamic Schema Changes ? Flexible querying mechanisms4/29/201162Data collection and analysis Ce.g. Collecting data from Social networking sitesDataImage DataComputational TechniquesImage Processing Human ComputingHealth KnowledgeSubstance Use (Drinking, Smoking, Drugs, etc)Comments Information Retrieval Techniques Friend Network Semantic Analysis Social Network based Games Friends Health Behavior Social Network Analysis Behavioral Data Mining Game Analytics Propensity to be influenced Sexual BehaviorReading activity on websites4/29/201163Social Ecological ModelSocietalSmoke-free workplace policyOrganizationalSmokefree schoolInterpersonalReduced exposure to smoking peersIntrapersonalSmoking isn’t that coolReduced community tolerance for cigarette smoking4/29/2011No tobacco sponsor-ship of youth eventsFewer friends begin smokingSmoking isn’t that cool64Challenges and Issues to incorporate Social Network Data? Technical challenges ? Data extraction? Social Network Aggregation: Gathering information from various socialnetworks and combining them at a single location ? De-duplication: Identifying a user from multiple sources as the same users? Image processing? The limitation of automatic image annotation to capture and understandhealth related behavior? Social network analysis? Scalability issues : the large scale datasets that we should analyze? Logistical and ethical issues ? Some sites have strict privacy controls, difficult to obtain data ? Data on publicly visible sites were not intended to be used for research purposes? Necessary to obtain consent? ? If so, how to obtain consent from a large, complete network? ? Parental vs. adolescent consent4/29/201165Conducting theoretical and applied research in this emerging field? Scientific Research C use publicly available datasets suchas UCI datasets? Scientific breadth ? Replicable by other researchers? Some publicly available datasets? UCI Machine Learning Repository ? KDD Cup 2008 -Siemens (Requires registration) ? MIT-BIH Arrhythmia Database ? ECML/PKDD discovery challenge dataset. ? Healthcare Cost and Utilization Project (H-CUP) ? HIV Prevention Trials Network - Vaccine Preparedness Study/Uninfected Protocol Cohort ? National Trauma Data Bank (NTDB) ? Behavioral Risk Factor Surveillance System (BRFSS) ? Link to National Public Health Data Sets(http://www-users.cs.umn.edu/~desikan/pakdd2011/datasets.html)4/29/2011 66Conducting research with high impact? High Impact Research C use private datasets fromcollaborations? Need industry collaboration ? Not easily replicable ? Scientific breadth and validity may not be to a great extent ? Basic research question may arise from collaboration? Develop right interactive partnership. What can beobtained?? Access to data, ? Understand what the problems are, and ? You have people with power to implement ideas and help evaluateeffectivenessHealthcare Industry Academia Collaboration4/29/2011CENTER FOR HEALTHCARE INNOVATION67Key References? [Ahmad09] Farzana Kabir Ahmad, Safaai Deris and Nor Hayati Othman, “Imperative Growing ? ? ? ???? ?Trends toward Applying Integrated Data in Breast Cancer Prognosis”, MASAUM Journal of Basic and Applied Sciences, Vol.1, No.1 August 2009 [Berwick97] DM Berwick, “The total customer relationship in health care: broadening the bandwidth”, Jt Comm J Qual Improv. ):245-50 [Bolton02] Richard J. Bolton and David J. Hand, “Statistical Fraud Detection: A Review”, Statistical Science 2002, Vol. 17, No. 3, 235C255 [Burroni2004] Burroni M, Corona R, Dell’Eva G, et al. Melanoma computer-aided diagnosis: reliability and feasibility study. Clin Cancer Res 1C1886. [Buczak09] Buczak, A.L.; Moniz, L.J.; Feighner, B.H.; Lombardo, J.S.; Mining electronic medical records for patient care patterns. IEEE Symposium on Computational Intelligence and Data Mining (CIDM), 2009 [Campbell2010] Kevin Campbell, N. Marcus Thygeson and Stuart Speedie. Exploration of Classification Techniques as a Treatment Decision Support Tool for Patients with Uterine F Proceedings of International Workshop on Data Mining for HealthCare Management, PAKDD-2010. [Catley06] Christina Catley, Monique Frize, C. Robin Walker, and Dorina C. Petriu, “Predicting HighRisk Preterm Birth Using Arti?cial Neural Networks”, IEEE TRANSACTIONS ON INFORMATION TECHNOLOGY IN BIOMEDICINE, VOL. 10, NO. 3, JULY 2006 [Cerrito07] P.B. Cerrito: Mining the Electronic Medical Record to Examine Physician Decisions, Studies in Computational Intelligence (SCI) 48, 113C126 (2007) [Chattopadhyay08] S. Chattopadhyay, P. Ray, H.S. Chen, M.B. Lee and H.C. Chiang, “Suicidal Risk Evaluation Using a Similarity-Based Classifier”, ADMA 2008, LNAI 5139, pp. 51C61, 2008684/29/2011Key References? [DeToledo09] Paula de Toledo, Pablo M. Rios, Agapito Ledezma, Araceli Sanchis, Jose F. Alen, and? ? ? ???Alfonso Lagares, “Predicting the Outcome of Patients With Subarachnoid Hemorrhage Using Machine Learning Techniques”, IEEE TRANSACTIONS ON INFORMATION TECHNOLOGY IN BIOMEDICINE, VOL. 13, NO. 5, SEPTEMBER 2009 [Doniger02] Scott Doniger, Thomas Hofmann, Joanne Yeh. Journal of Computational Biology. December ): 849-864 [Gulera05] Inan Gulera, Elif Derya Ubeyli, “ECG beat classi?er designed by combined neural network model”, Pattern Recognition 38 ( C 208 [Hardin2008] J. Michael Hardin and David C. Chhieng, Clinical Decision Support Systems: Theory and Practice, Eta S. Berner (Editor), Springer Verlag, Health Informatics Series,44-63; 2008. [Hadzic2010] Maja Hadzic, Fedja Hadzic and Tharam Dillon et al. Mining of patient data: towards better treatment strategies for depression. International Journal of Functional Informatics and Personalised Medicine, 2010 [Hub2000] Huber S, Medl M,Vesely M, Czembirek H, Zuna I, Delorme S. Ultrasonographic tissue characterization in monitoring tumor response to neoadjuvant chemotherapy in locally advanced breast cancer. J Ultrasound Med C686. [Hibbard1997] Hibbard LS, McKeel DW Jr. Automated identification and quantitative morphometry of the senile plaques of Alzheimer’s disease. Anal Quant Cytol Histol C138.4/29/201169Key References? [Jiang06] Zheng Jiang, Kazunobu Yamauchi, Kentaro Yoshioka, Kazuma Aoki, Susumu Kuroyanagi, ? ????Akira Iwata, Jun Yang, and Kai Wang. 2006. Support Vector Machine-Based Feature Selection for Classification of Liver Fibrosis Grade in Chronic Hepatitis C. J. Med. Syst. 30, 5 (October 2006), 389-394 [Koh05] Koh HC, Tan G. “ Data mining applications in healthcare”. J Healthc Inf Manag. ):64-72. [Kohli01] Rajiv Kohli, Frank Piontek, Tim Ellington, Tom VanOsdol, Marylou Shepard, Gary Brazel, “Managing customer relationships through E-business decision support applications: a case of hospital-physician collaboration”, Decision Support Systems, Volume 32, Issue 2, December 2001, Pages 171-187 [Kuttikrishnan10] Murugesan Kuttikrishnan, Indumathi Jeyaraman, Manjula Dhanabalachandran, “An Optimised Intellectual Agent Based Secure Decision System for Health Care”, International Journal of Engineering Science and Technology, Vol. 2(8), -3675 [Kwiatkowska07] Mila Kwiatkowska, M. Stella Atkins, Najib T. Ayas, and C. Frank Ryan, “Knowledge-Based Data Analysis: First Step Toward the Creation of Clinical Prediction Rules Using a New Typicality Measure”, IEEE TRANSACTIONS ON INFORMATION TECHNOLOGY IN BIOMEDICINE, VOL. 11, NO. 6, NOVEMBER 2007 [Laxminarayan06] Parameshvyas Laxminarayan, Sergio A. Alvarez, Carolina Ruiz, and Majaz Moonis, “Mining Statistically Signi?cant Associations for Exploratory Analysis of Human Sleep Data”, IEEE TRANSACTIONS ON INFORMATION TECHNOLOGY IN BIOMEDICINE, VOL. 10, NO. 3, JULY 2006704/29/2011Key References? [Ludwick09] Ludwick DA, Doucette J. Adopting electronic medical records in primary care: lessons ? ? ???? ?learned from health information systems implementation experience in seven countries. Int J Med Inform. ):22-31 [Maglogiannis09] Ilias Maglogiannis, Introducing Intelligence in Electronic Healthcare Systems: State of the Art and Future Trends. Artificial Intelligence, LNAI 5640, pp. 71 C 90, 2009 [McGarry07] Ken McGarry, James Chambers, and Giles Oatley. 2007. A multi-layered approach to protein data integration for diabetes research. Artif. Intell. Med. 41, 2 (October 2007), 129-14 [O’Sullivan08] Dympna O’Sullivan, William Elazmeh, Szymon Wilk, Ken Farion, Stan Matwin, Wojtek Michalowski, and Morvarid Sehatkar, “Using Secondary Knowledge to Support Decision Tree Classi?cation of Retrospective Clinical Data”, MCD 2007, LNAI 4944, pp. 238C251, 2008 [Ouschan06] Robyn Ouschan, Jillian Sweeney, Lester Johnson, “Customer empowerment and relationship outcomes in healthcare consultations”, European Journal of Marketing, Vol. 40 Iss: 9/10, pp.1068 C
[Palaniappan08] Palaniappan, S.; Awang, R., “Intelligent heart disease prediction system using data mining techniques”, IEEE/ACS International Conference on Computer Systems and Applications, 2008. AICCSA 2008 [Penny09] Kay Penny and Thomas Chesney, “Mining Trauma Injury Data with Imputed Values”, Statistical Analysis and Data Mining 2: 246 C 254, 2009 [Persson09] Persson, M.; Lavesson, N., “Identification of Surgery Indicators by Mining Hospital Data: A Preliminary Study”, 20th International Workshop on Database and Expert Systems Application, 2009. DEXA &#39;09714/29/2011Key References? [Puschmann01] Puschmann, T.; Alt, R., “Customer relationship management in the pharmaceutical ? ? ?? ??industry”, Proceedings of the 34th Annual Hawaii International Conference on System Sciences, 2001 [Razavi07] Amir R. Razavi & Hans Gill & Hans Ahlfeldt, Nosrat Shahsavar, “Predicting Metastasis in Breast Cancer: Comparing a Decision Tree with Domain Experts”, J Med Syst (C273 [Rygielski02] Chris Rygielski, Jyun-Cheng Wang, David C. Yen, “Data mining techniques for customer relationship management”, Technology in Society 24 (2 [Santos09] M. F. Santos, F. Portela, M. Vilas-Boas, J. Machado, A. Abelha, J. Neves, A. Silva, F. Rua, M. Salazar, C. Quintas, and A. F. Cabral. 2009. Nursing information architecture for situated decision support in intensive care units. In Proceedings of the 9th WSEAS international conference on Applied informatics and communications (AIC&#39;09), Nikos E. Mastorakis, Metin Demiralp, Valeri Mladenov, and Zoran Bojkovic (Eds.). World Scientific and Engineering Academy and Society (WSEAS), Stevens Point, Wisconsin, USA, 491-496 [Stolba06] N. Stolba, A. Tjoa: &The Relevance of Data Warehousing and Data Mining in the Field of Evidence-Based Medicine to Support Healthcare Decision Making&; ICCS 2006 [Young00] Young, Gary J.; Meterko, M Desai, Kamal R., “Patient Satisfaction With Hospital Care: Effects of Demographic and Institutional Characteristics”, Medical Care: March 2000 - Volume 38 - Issue 3 - pp 325-334 [Zellner2004] Zellner BB, Rand SD, Prost R, Krouwer H, Chetty VK.A cost-minimizing diagnostic methodology for discrimination between neoplastic and non-neoplastic brain lesions: utilizing a genetic algorithm. Acad Radiol C177724/29/2011
/view/314333.htm 2011-9-26 - ...and healthcare 3)文档详情主题 AAppointments...for construction project management in university ...翻译模板2011翻译模板2011隐藏&& 文献翻译 使用正确的文献题录格式: 责任人, 题目...provides a generalized platform for data mining topics on the internet, from...Data mining software is one of a number of analytical tools for analyzing ...Data warehousing is defined as a process of centralized data management and ...Microsoft SQL Server 2008 Data Mining Add-ins for Microsoft Office 2007, ...2008 Management Studio Express Microsoft SQL Server 2005 Express Edition ...Management, Shanghai University, Shanghai 200444 Abstract: Decision Support System (DSS) is an important application of data mining in the enterprise ...mid term course work for data mining_IT/计算机_专业资料。300个stock 高频数据的k-means聚类 Part1:Codes # An open comparative study on high-frequency #...Workflow Management S Business Process Reengineering...; Data Mining and Knowledge D Knowledge ...Intelligent Healthcare S E-learning and In...COMP5318 Knowledge Discovery and Data Mining_2011 Semester 1_tutorialweek11_...4. What&#39;s the formula for Bayes&#39; theorem and why is it useful? 5. ...Management (ERP) [3], Customer Relationship Management (CRM)[4], and Web...research specifically for web data mining software to tap the process becomes...Data Mining常用词汇表-中英文_互联网_IT/计算机_专业资料。1R 算法名称 Activation...Data warehouse 数据仓库 Database (DB) 数据库 Database Management System 数...
All rights reserved Powered by
copyright &copyright 。文档资料库内容来自网络，如有侵犯请联系客服。}