AKash ISI Presentation 101
-
Upload
akash-mavle -
Category
Documents
-
view
15 -
download
0
description
Transcript of AKash ISI Presentation 101
Predictive Analytics using Watson
1
Session Agenda
• Introduction to Watson and IBM Content and Predictive Analytics for Healthcare (ICPA)
• Using ICPA to Prevent CHF Readmissions– Seton Healthcare Family Case Study
2
PrimeNumerics Profile• PrimeNumerics Consulting(www.primenumerics.com) is a division of
Reporting House in Collaboration with Solversa Technologies with core focus on ERP consulting for Infor, SAP and Epicor
• Reporting House Inc.(www.reportinghouse.com) is 18 year old company specialized in Analytics, Performance Management, Reporting and Business Intelligence with offices in NC, US and Pune, India
• Solversa Technologies (I) Pvt Ltd ( www.solversa.com ) is a 10 year old company specialized in Baan/Infor consulting
Consulting Services
Analytics Driven Manufacturing
• JIT, Toyota Production Systems, TQM, TPM, PokaYoke, 5S, Value Stream Mapping, Lean Manufacturing
• Manufacturing Performance Measurement
Outsourced CIO •ERP SRS Creation, ERP Evaluation and Implementation, ERP Project Oversight Consulting•IT Strategy Formulation•IT Portfolio Management
Implementation Services •Baan/Infor Implementation, Customization, Development•Baan/Infor offsite support•ERP optimization
Our Global Clients
Other Clients
Extend your analytics capabilities where you want to be…
Sense & Respond Predict & Act
RawData
CleanedData
Standard Reports
Ad Hoc Reports &
OLAP
Generic Predictive Analytics
Predictive Modeling
Optimization
What happened?
Why did it happen?
What will happen?
What is the best that could happen?
Com
petiti
ve A
dvan
tage
Analytics Maturity
Degree of Intelligence
Standard Reports
Ad Hoc Reports
Query/Drill Down
KPIs/Alerts
What happened?
How many, how often, where?
Where exactly is the problem?
What actions are needed?
Rear
Vie
w
Statistical Analysis
Forecasting/Extrapolation
Predictive Modeling
Optimization
Why is this happening?
What if these trends continue?
What will happen next?
What is the best that can happen?
Forw
ard
View
A Non-Homogeneous Poisson Process predictive model
for Automobile Warranty Claims
1. Introduction:
We know that the current trends of marketing of automobiles are increasing day by day. Therefore, the need for predicting the time of failure of vehicles and thus the warranty claims is also increasing. The present paper includes the comparison of methods of predicting warranty claims which will be then useful to save the recourses for paying the future warranty claims. Also the paper shows the comparative use of non-homogeneous Poisson process(NHPP) predictive model for vehicles.
. . .
Characterization of warranties:
It depends on manufacturer that what type of warranty on which item is provide to the buyer of product.
A warranty can be provided which will be renewed at the time of first failure of a product which is considered to be a renewable warranty. Another is non renewable warranty which has a fixed duration of warranty coverage from the time of sale.
Warranty is also classified on the basis of number of variables affecting the warranty. Thus the types are:1. One -dimensional warranty which depends on one
variable (time since purchase).2. Two-dimensional warranty which depends on two
variables (time and miles covered)
. . .
• The one dimensional warranty model exist for both renewing and non renewing free replacement models. Here we will discuss about two- dimensional warranty.
• In two dimensional warranty the vehicles which have crossed the mileage limit are censored from the total population of vehicles i.e. now they are out of warranty coverage and the remaining number of vehicles is considered for prediction of warranty claims
• Consider for two dimensional the time limit and the mileage limit as tWL and mWL, respectively. satisfy the following If we have a usage function U(t) which relates to time and mileage then vehicles which are repaired at first failure are eligible for warranty claims if they
Two-dimensional warranty coverage in the time and mileage domain
Automobile warranty predictive model
• A NHPP predictive model for automobile warranty claims that provides a great deal of flexibility when fitting observational data. The approach presented here makes predictions in the time domain but accounts for mileage by removing vehicles from the population when they exceed the mileage limit.
• Nm (t) a population size that quantifies the number at risk in the time domain and takes into account the effects of mileage by removing vehicles when they exceed the mileage limit.
Time-based warranty claim rate
• The NHPP is characterized by the intensity function that represents the instantaneous probability of a failure.
Predicting warranty claims with the model
• The function is predicted number of claims through time t as the expected value of the cumulative warranty claim function
Where function C(t) represent the cumulative number of warranty claims observed by the population through t time in service.
Estimating the number at risk
• To estimate the number at risk function Nm(t) for the prediction period, a usage value u was calculated for each of vehicles with an observed mileage value prior to the cutoff date.
• A histogram of the usage values follows Gamma distribution with probability density function
Usage Data
Estimating the NHPP intensity function
• Subsystem 1:The Weibull-uniform mixture model with cumulative
distribution function
Estimating the NHPP intensity function• Subsystem 2:Weibull distribution to model field failure with
cumulative distribution function
Subsystem 3:Linear hazard function
Conclusion• Automobile manufacturers rely on predictions for the number, timing, and
cost of future warranty claims.• Omitting the mileage data contained on automobile warranty claims, and
making predictions purely in the time domain, ignores vehicles leaving coverage prior to the warranty time limit.
• Incorporating a two-dimensional aspect will result in a predictive model that better represents the process being modeled.
• A non-homogenous Poisson process (NHPP) prédictive model that has a parametric component (time to first failure) and provides a great deal of flexibility in application.
• The NHPP approach also allows incorporating past experience when identifying the failure process and bases predictions on early field performance of the products.
• Using the NHPP will result in more accurate predictions of warranty claims and support decision making when implementing engineering design and manufacturing process changes.
• More accurate predictions will also assist the manufacturer in allocating reserves to pay for repairs covered under warranty.
FORECASTING REPAIR COST
•We have to forecast the Repair Cost•Our Data contain failure date, failure code, repair cost ,material cost, labor cost etc.•In this data the dependent variable is repair cost and other are independent variable i.e. material cost ,labor cost ,failure code ,tool cost etc.
We use two methods Generalized Regression model and Neural network model for predicting the Repair Cost.
Conclusion
• We conclude that Regression Method is better than Neural Network for predicting the Repair Cost.
Inconsistent quality and increasing costs require healthcare transformation in key areas
37 * New England Journal of Medicine – Rehospitalization Among Patients in the Medicare Fee-for-Service Program, April 2009** http://www.healthleadersmedia.com/content/COM-263665/3-Readmissions-to-Reduce-Now
The need for better clinical outcomes
• One in five patients suffer from preventable readmissions … represents $17.4 billion of the current $102.6 billion Medicare budget*
• 1.5 million patients in the U.S. harmed annually by errors in the way medications are prescribed, delivered and taken
The need for better operational outcomes
In 2012, Hospitals will be penalized for high readmission rates - Medicare discharge payments starting will be reduced in key areas**
• $475 billion: Estimated annual US healthcare spending on administrative and clinical waste, fraud, abuse and other waste
With this change comes an opportunity to exploit the explosion of information
38
… yet some health organizations operate with blind spots and information is not actionable
Volume of information → Lack of Insight
1 in 3 managers frequently make critical decisions without the information they need
Variety of information → Inefficient Access
1 in 2 don’t have access to the information across their organization needed to do their jobs … notably unstructured information including paper
Velocity of decision making → Inability to Predict
3 in 4 business leaders say more predictive information would drive better decisions
15 petabytesAmount of new information created each day - eight times more than the information in all US libraries1
Health data growing 35% per
year*
* Recent study by Enterprise Strategy Group
39
In May 1898 Portugal celebrated the 400th anniversary of this
explorer’s arrival in India
Remember to Answer in the Form of a Question
© 2014 PrimeNumerics
explorer
India
In May 1898
India
In May
celebrated
anniversary
in Portugal
In May, Gary arrived in India after he celebrated his anniversary in
Portugal
Portugal
400th anniversary
celebrated
Gary
40
In May 1898 Portugal celebrated the 400th anniversary of this explorer’s
arrival in India
This evidence suggests “Gary” is the answer
BUT the system must learn that keyword
matching may be weak relative to other types of
evidence
arrived in
arrival in
Legend
Keyword “Hit”
Reference Text
Answer
Weak evidenceRed Text
Why is Jeopardy! so difficult?Answering complex natural language questions requires more than keyword evidence
© 2014 PrimeNumerics
27th May 1498
Vasco da Gama
landed in
arrival in
explorer
India
Para-phrases
Geo-KB
DateMatch
41
Stronger evidence can be much harder to find
and score …
… and the evidence is still not 100% certain
Search far and wide
Explore many hypotheses
Find judge evidence
Many inference algorithms
On the 27th of May 1498, Vasco da Gama landed in Kappad Beach
400th anniversary
Portugal
May 1898
celebrated
In May 1898 Portugal celebrated the 400th anniversary of this explorer’s arrival
in India.
Kappad Beach
Legend
Temporal Reasoning
Reference Text
Answer
Statistical Paraphrasing
GeoSpatial Reasoning
Watson Leverages Multiple Algorithms to Gather Deeper Evidence
• How are you measuring and reducing preventative readmissions?
• How are you providing clinicians with targeted diagnostic assistance?
• Which patients are following discharge instructions?
• How are you leveraging unstructured data to prevent and detect fraud?
• How are you using data to predict intervention program candidates?
• Would revealing insights trapped in unstructured information facilitate more informed decision making?
... but the biggest blind spot still remains
Physician notes and discharge summaries Patient history, symptoms and non-symptoms Pathology reports Tweets, text messages and online forums Satisfaction surveys Claims and case management data Forms based data and comments Emails and correspondence Trusted reference journals including portals Paper based records and documents
42 * AIIM website, accepted industry percentage
Over 80% of stored health information is unstructured*
Does unlocking the unstructured data help accelerate your transformation?
Medical Transcription Discharge Summary Sample # 2:DATE OF ADMISSION: MM/DD/YYYYDATE OF DISCHARGE: MM/DD/YYYY ADMITTING DIAGNOSIS: Syncope.CHIEF COMPLAINT: Vertigo or dizziness.HISTORY OF PRESENT ILLNESS: This is an (XX)-year-old male with a past medical history of coronary artery disease, CABG done a few years ago, atrial fibrillation, peripheral arterial disease, peripheral neuropathy, recently retired one year ago secondary to leg pain. The patient came to the ER for an episode of vertigo while reaching for some books. The patient was able to reach the books, to support self, but did not have any syncope. No nausea or vomiting. No chest pain. No shortness of breath. Came to ER and had a CT head, which was within normal limits. The impression was atrophy with old ischemic changes but no acute intracranial findings. No focal weakness, headache, vision changes or speech changes. The patient has had similar episodes since one year. Peripheral neuropathy since one year and not relieved with multiple medications. The patient also complains of weight loss of 25 pounds in the last 6 months. No colonoscopy done. Recent history of hematochezia but believes it was secondary to proctitis and secondary to decreased appetite. No nausea, vomiting, no abdominal pain.PROCEDURES PERFORMED: The patient had a chest x-ray, which showed cardiomegaly with atherosclerotic heart disease, pleural thickening and small pleural effusion, a left costophrenic angle which has not changed when compared to prior examination, COPD pattern. The patient also had a head CT which showed atrophy with old ischemic changes. No acute intracranial findings.CONSULTS OBTAINED: A rehab consult was done.PAST MEDICAL/SURGICAL HISTORY: Positive for atrial fibrillation. The patient had AVR 6 years ago. Peripheral arterial disease with hypertension, peripheral neuropathy, atherosclerosis, hemorrhoids, proctitis, CABG, and cholecystectomy.FAMILY HISTORY: Positive for atherosclerosis, hypertension, autoimmune diseases in the family.SOCIAL HISTORY: Never smoked. Alcohol socially. No drugs.ALLERGIES: NO KNOWN DRUG ALLERGIES.REVIEW OF SYMPTOMS: Weight loss of 25 pounds within the last 6 months, shortness of breath, constipation, bleeding from hemorrhoids, increased frequency of urination, muscle aches, dizziness and faintness, focal weakness and numbness in both legs, knees and feet.PHYSICAL EXAMINATION: VITAL SIGNS: Blood pressure 188/74, pulse 62, respirations 18 and saturation of 98% on room air. General Appearance: The patient is a pleasant man, comfortable. HEENT: Conjunctivae are normal. PERRLA. EOMI. NECK: No masses. Trachea is central. No thyromegaly. LUNGS: Clear to auscultation and percussion bilaterally. HEART: Irregular rhythm. ABDOMEN: Soft, nontender, and nondistended. Bowel sounds are positive. GENITOURINARY: Prostate is hypertrophic with smooth margin. EXTREMITIES: Upper and lower limbs bilaterally normal. SKIN: Normal. NEUROLOGIC: Cranial nerves are grossly within normal limits. No nystagmus. DTRs are normal. Good sensation. The patient is alert, awake, and oriented x3. Mild confusion.LABORATORY DATA AND RADIOLOGICAL RESULTS: WBC 8.6, hemoglobin 13.4, hematocrit 39.8, platelets 207,000, MCV 91.6, neutrophil percentage of 72.6%. Sodium 133, potassium 4.7, chloride 104. Blood urea nitrogen of 18 and creatinine of 1.1. PT 17.4, INR 1.6, PTT 33.The patient had a chest x-ray, which showed cardiomegaly with atherosclerotic heart disease, pleural thickening and small pleural effusion, a left costophrenic angle which has not changed when compared to prior examination, COPD pattern. The patient also had a head CT, which showed atrophy with old ischemic changes. No acute intracranial findings.HOSPITAL COURSE AND TREATMENT: This is an (XX)-year-old male with syncope.1. Syncope. This may be secondary to questionable cerebral ischemia/atrial fibrillation/hypotension, so Neurology was kept on board and the patient was scheduled for a carotid Doppler and a 2-D echo. Orthostatics were ordered. Vitamin B12, TSH, free T4 and T3 were ordered along with cortisol level in the morning. FOBT x3 were done and cardiology followup as outpatient. The patient had a carotid Doppler done on the next day and it showed mild irregular plaque disease, right and left internal carotid arteries, approximately 20-59%. The patient's vitamin B12 level came the next morning and the level was 1180. His folate was 18.7 and his TSH was 1.98, free T4 of 1.38 and T4 level of 7.4, cortisol level of 15.4, which are within normal limits. Dr. Doe, who is the patient's cardiologist, was informed. Dr. Doe was kind enough to see the patient the very next day, and his impression was that the patient has atrial fibrillation, rate controlled, status post AVR, St. Jude, and peripheral neuropathy. Subtherapeutic INR, the patient's relative target INR is 2-3. He suggested PT evaluation and suggested a low dose of SSRI and Dr. Doe was of the opinion that the patient does not need any further cardiac recommendation. CT chest, abdomen, and pelvis were done. CT chest had an impression of coronary artery calcification, aortic valve replacement, cardiomegaly, suspect a very small left pleural effusion, no acute active pulmonary disease. CT abdomen and pelvis showed prior cholecystectomy, diverticulosis of sigmoid colon, two benign-appearing simple cysts involving the right kidney, calcified arteriosclerotic plaque disease of the abdominal aorta and iliac vessels bilaterally. The patient was ruled out of any malignancy whatsoever.2. Hypertension. The patient at home was on Cardizem ER 90 mg thrice daily, and it was changed initially to Cardizem 90 mg thrice daily, and then with Dr. Doe's request, we changed the Cardizem to 240 mg t.i.d.3. Atrial fibrillation with subtherapeutic INR. The patient at home was on Digitalis. That was continued. Dr. Doe was of the opinion that the patient himself takes care of the Coumadin, and Dr. Doe was of the opinion that probably that is why the patient is not able to maintain therapeutic INR. In the hospital, the patient's warfarin was increased to 5 mg q.h.s., and at the time of the discharge, he was requested to follow his appointments so that his INR can be maintained.4. Gout. The patient was on allopurinol. There were no acute issues regarding the gout.5. Prophylaxis. The patient was on Protonix and TEDs.6. Social. The patient is FULL CODE.
DISCHARGE DIAGNOSIS: Syncope.
DISCHARGE DISPOSITION: The patient is discharged to home.
DISCHARGE MEDICATIONS: The patient was discharged on the following medications; Cardizem 90 mg p.o. thrice daily, digoxin 0.125 mg p.o. once daily, allopurinol 100 mg two times daily, Coumadin 4 mg p.o. q.h.s., and Remeron 15 mg p.o. q.h.s.
DISCHARGE INSTRUCTIONS: Since the patient had generalized deconditioning, the patient was advised home PT, OT and that was arranged for the patient.
DISCHARGE DIET: Cardiac diet.
DISCHARGE ACTIVITY: Resume activity as tolerated.
Echocardiogram Sample Report:DATE OF STUDY: MM/DD/YYYYDATE OF INTERPRETATION OF STUDY: Echocardiogram was obtained for assessment of left ventricular function. The patient has been admitted with diagnosis of syncope. Overall, the study was suboptimal due to poor sonic window.FINDINGS:1. Aortic root appears normal.2. Left atrium is mildly dilated. No gross intraluminal pathology is recognized, although subtle abnormalities could not be excluded. Right atrium is of normal dimension.3. There is echo dropout of the interatrial septum. Atrial septal defects could not be excluded.4. Right and left ventricles are normal in internal dimension. Overall left ventricular systolic function appears to be normal. Eyeball ejection fraction is around 55%. Again, due to poor sonic window, wall motion abnormalities in the distribution of lateral and apical wall could not be excluded.5. Aortic valve is sclerotic with normal excursion. Color flow imaging and Doppler study demonstrates trace aortic regurgitation.6. Mitral valve leaflets are also sclerotic with normal excursion. Color flow imaging and Doppler study demonstrates trace to mild degree of mitral regurgitation.7. Tricuspid valve is delicate and opens normally. Pulmonic valve is not clearly seen. No evidence of pericardial effusion. CONCLUSIONS:1. Poor quality study.2. Eyeball ejection fraction is 55%.3. Trace to mild degree of mitral regurgitation.4. Trace aortic regurgitation.
Cardiology Consultation Transcribed Medical Transcription Sample ReportsREFERRING PHYSICIAN: John Doe, MD CONSULTING PHYSICIAN: Jane Doe, MD HISTORY OF PRESENT ILLNESS: This (XX)-year-old lady is seen in consultation for Dr. John Doe. She has been under consideration for ventral hernia repair and has a background of aortic valve replacement and known coronary artery disease. The patient was admitted with complaints of abdominal pain, anorexia, and vomiting. She underwent a CT scan of the abdomen and pelvis and this showed the ventral hernia involving the transverse colon, but without strangulation. There was an atrophic right kidney. She had bilateral renal cysts. The hepatic flexure wall was thickened. There was sigmoid diverticulosis without diverticulitis. It has been recommended to her that she undergo repair of the ventral hernia. For this reason, cardiology consult is obtained to assess whether she can be cared from the cardiac standpoint.PAST CARDIAC HISTORY: Bypass surgery. She underwent echocardiography and cardiac catheterization prior to the operation. Echocardiography showed an ejection fraction of 50%. There was marked left ventricular hypertrophy with septal wall 1.60 cm and posterior wall 1.55 cm. Coronary arteriography showed 90% stenosis in the anterior descending artery, situated distally just before the apex of the left ventricle. Only mild to moderate narrowing was seen elsewhere in the coronary circulation.CORONARY RISK FACTORS: Her father had an irregular heartbeat and her brother had a fatal heart attack. She herself has had high blood pressure for 20 years. She has elevated cholesterol and takes Lipitor. She has had diabetes for 20 years. She is not a cigarette smoker. She does little physical exercise.REVIEW OF SYMPTOMS: CARDIOVASCULAR AND RESPIRATORY: She has no chest pain. She sometimes becomes short of breath if she walks too far. No cough. She has occasional swelling of her feet. Occasionally, she gets mildly lightheaded. Has not lost consciousness. She tends to be aware of her heartbeat when she is tired. She has no history of heart murmur or rheumatic fever. GASTROINTESTINAL: Recent GI symptoms as noted above, but she does not usually have such problems. She has had no hematemesis. She has no history of ulcer or jaundice. She sometimes has loose stools. No constipation and no blood in the stool. GENITOURINARY: She tends to have urinary frequency. She gets up once at night to pass urine. No dysuria, incontinence. She has had previous urinary infections. No stones noted. NEUROLOGIC: She has occasional headaches. No seizures. No trouble with vision, hearing, or speech. No limb weakness. MUSCULOSKELETAL: She tends to have joint and muscle pains and has a history of gout. HEMATOLOGIC: No anemia, abnormal bleeding, or previous blood transfusion. GYNECOLOGIC: No gynecologic or breast problems.PAST MEDICAL HISTORY: She has had shoulder and hand injuries and has had carpal tunnel surgery. She has been diabetic and has been on insulin. She has chronic renal insufficiency with creatinine around 2.2. She has had hypothyroidism. She has had morbid obesity. She has chronic obstructive sleep apnea and uses BiPAP. She has had hysterectomy and oophorectomy in the past. Otherwise as noted above.MEDICATIONS: Prior to hospital, she was taking glipizide XL 2.5 mg daily, metoprolol 50 mg b.i.d., Cipro 250 mg b.i.d., atorvastatin 40 mg daily, Synthroid 75 mcg daily, aspirin 81 mg daily, and Lantus 36 units daily. Currently, she is taking Lipitor 40 mg daily, Lantus 10 units at bedtime, Synthroid 75 mcg daily, metoprolol 50 mg b.i.d., and Zosyn 2.25 grams q.6h.SOCIAL HISTORY: She does not drink alcohol.PHYSICAL EXAMINATION:GENERAL APPEARANCE: She is not currently dyspneic, in no distress. She is alert, oriented, and pleasant.HEENT: Pupils are normal and react normally. No icterus. Mucous membranes well colored.NECK: Supple. No lymphadenopathy. Jugular venous pressure not elevated. Carotids equal. HEART: The heart rate is 82 per minute and regular and the blood pressure 132/78. The cardiac impulse has a normal quality. There is a grade 3/6 ejection systolic murmur heard medial to the apex and at the aortic area, with well heard radiation to the neck vessels.CHEST: Chest is clear to percussion and auscultation. Normal respiratory effort.ABDOMEN: Soft and nontender. The presence of a large ventral hernia is noted.EXTREMITIES: There is no edema. Posterior tibial pulses were felt bilaterally, but I did not feel the dorsalis pedis.SKIN: No rash or significant lesions are noted.LABORATORY AND DIAGNOSTIC DATA: Electrolytes are normal. BUN and creatinine 18/2.2. Blood sugar 150. White count is 7.6, hemoglobin 11.7 with hematocrit 34.9, platelets 187,000. LFTs were normal. Hemoglobin A1c 7.7. TSH 1.82. Troponin I was normal on three occasions.Chest x-ray showed an enlarged heart with postoperative changes, but no evidence of acute pathology. EKG shows probable left atrial enlargement. Low voltage QRS, probable inferior wall myocardial infarction and anterior wall infarction, age undetermined.ASSESSMENT:1. Aortic valve replacement with bioprosthetic valve. Residual systolic murmur.2. Arteriosclerotic heart disease with severe stenosis in anterior descending artery, but this is situated distally and subtends only a small mass of myocardium.3. Well preserved left ventricular systolic function. The EKG appearance of previous myocardial infarction is probably serious, indicating multiple other medical problems as listed above and also documented in the chart.RECOMMENDATIONS: It appears that she does not wish to proceed with the surgery at this time, and if such surgery is not
Cardiology Consultation Transcribed Medical Transcription Sample ReportsDATE OF CONSULTATION: MM/DD/YYYYREFERRING PHYSICIAN: John Doe, MD CONSULTING PHYSICIAN: Jane Doe, MDREASON FOR CONSULTATION: Surgical evaluation for coronary artery disease. HISTORY OF PRESENT ILLNESS: The patient is a (XX)-year-old female who has a known history of coronary artery disease. She underwent previous PTCA and stenting procedures in December and most recently in August. Since that time, she has been relatively stable with medical management. However, in the past several weeks, she started to notice some exertional dyspnea with chest pain. For the most part, the pain subsides with rest. For this reason, she was re-evaluated with a cardiac catheterization. This demonstrated 3-vessel coronary artery disease with a 70% lesion to the right coronary artery; this was a proximal lesion. The left main had a 70% stenosis. The circumflex also had a 99% stenosis. Overall left ventricular function was mildly reduced with an ejection fraction of about 45%. The left ventriculogram did note some apical hypokinesis. In view of these findings, surgical consultation was requested and the patient was seen and evaluated by Dr. Doe. PAST MEDICAL HISTORY: 1. Coronary artery disease as described above with previous PTCA and stenting procedures. 2. Dyslipidemia.3. Hypertension.4. Status post breast lumpectomy for cancer with followup radiation therapy to the chest. ALLERGIES: None. MEDICATIONS: Aspirin 81 mg daily, Plavix 75 mg daily, Altace 2.5 mg daily, metoprolol 50 mg b.i.d. and Lipitor 10 mg q.h.s.SOCIAL HISTORY: She quit smoking approximately 8 months ago. Prior to that time, she had about a 35- to 40-pack-year history. She does not abuse alcohol. FAMILY MEDICAL HISTORY: Mother died prematurely of breast cancer. Her father died prematurely of gastric carcinoma. REVIEW OF SYMPTOMS: There is no history of any CVAs, TIAs or seizures. No chronic headaches. No asthma, TB, hemoptysis or productive cough. There is no congenital heart abnormality or rheumatic fever history. She has no palpitations. She notes no nausea, vomiting, constipation, diarrhea, but immediately prior to admission, she did develop some diffuse abdominal discomfort. She says that since then, this has resolved. No diabetes or thyroid problem. There is no depression or psychiatric problems. There is no musculoskeletal disorders or history of gout. There are no hematologic problems or blood dyscrasias. No bleeding tendencies. Again, she had a history of breast cancer and underwent lumpectomy procedures for this with followup radiation therapy. She has been followed in the past 10 years and mammography shows no evidence of any recurrent problems. There is no recent fevers, malaise, changes in appetite or changes in weight. PHYSICAL EXAMINATION: Her blood pressure is 120/70, pulse is 80. She is in a sinus rhythm on the EKG monitor. Respirations are 18 and unlabored. Temperature is 98.2 degrees Fahrenheit. She weighs 160 pounds, she is 5 feet 4 inches. In general, this was an elderly-appearing, pleasant female who currently is not in acute distress. Skin color and turgor are good. Pupils were equal and reactive to light. Conjunctivae clear. Throat is benign. Mucosa was moist and noncyanotic. Neck veins not distended at 90 degrees. Carotids had 2+ upstrokes bilaterally without bruits. No lymphadenopathy was appreciated. Chest had a normal AP diameter. The lungs were clear in the apices and bases, no wheezing or egophony appreciated. The heart had a normal S1, S2. No murmurs, clicks or gallops. The abdomen was soft, nontender, nondistended. Good bowel sounds present. No hepatosplenomegaly was appreciated. No pulsatile masses were felt. No abdominal bruits were heard. Her pulses are 2+ and equal bilaterally in the upper and lower extremities. No clubbing is appreciated. She is oriented x3. Demonstrated a good amount of strength in the upper and lower extremities. Face was symmetrical. She had a normal gait.IMPRESSION: This is a (XX)-year-old female with significant multivessel coronary artery disease. The patient also has a left main lesion. She has undergone several PTCA and stenting procedures within the last year to year and a half. At this point, in order to reduce the risk of any possible ischemia in the future, surgical myocardial revascularization is recommended. PLAN: We will plan to proceed with surgical myocardial revascularization. The risks and benefits of this procedure were explained to the patient. All questions pertaining to this procedure were answered.
Unstructured Data is Messy but Filled with Key Medical Facts
Medications, Diseases, Symptoms, Non-Symptoms, Lab Measurements, Social History, Family History and Much More
Claims Analysis
Unstructured Information Use LandscapeNatural Language Processing is Needed to
Commerce Search
Search Analyze / Visualize(Trends, Patterns, Relationships)
Deep QA
Safety, Defects,
Maintenance
Enterprise Search
eDiscovery, Legal
Risk, Fraud,
Security
Semantic Understanding, Ontology Mgt and Big Data
VoC, Churn,
Cust Svc
Social Media,
Marketing
Product Quality
Expertise Locator
Knowledge Mgt
Call Center,
Help, Self Service
Research (Biz, Edu, Legal,
Scientific)
Trading Support
Healthcare Operational
Healthcare Clinical
Publishing (Tag, Locate)
Iterative Q&A
Incident Management
"We anticipate this solution to be a game changer in biomedical research and patient care … accelerate the pace of clinical and translational research …"
Dr. Rakesh Nagarajan, MD, PhD, Associate Professor, Department of Pathology and
Immunology, Washington University
45
Business ChallengeExisting Biomedical Informatics (BMI) resources were disjointed, siloed, redundant and only available to a few researchers - key insights not accessible, trapped in unstructured clinical notes, diagnostic reports, etc.
What’s Smart?Leverage unstructured information along with structured data by using IBM Content Analytics with IBM InfoSphere Warehouse
Smarter Business OutcomesResearchers now able to see new trends, patterns and find answers in days instead of weeks or months eliminating manual methods also enables new grant revenue
BJC Healthcare and Washington University PartnershipImproving care and increasing revenue while lowering research costs
IBM is helping to transform healthcareRevealing clinical and operational insights in the high impact overlap between clinical and operational – enabling low cost accountable care
46
Diagnostic assistance
Clinical treatment effectiveness
Critical care intervention
Research for improved disease management
Readmission prevention
Claims management
Fraud detection and prevention
Voice of the patient
Patient discharge and follow-up care
IBM Content and Predictive Analytics for Healthcare
Improved patient satisfaction at lower costs Enhanced patient care with optimized
outcomes
ClinicalOutcome
s
Operational
Outcomes
47
Unstructured Data(Nurses notes, discharge notes, etc.)
Structured Data(Billing data, EMR, etc.)
Raw Information
Search and Visually Explore (Mine)
Monitor, Dashboard and Report
Question and Answer*
Custom Solutions
Dynamic MultimodeInteraction
Content and Predictive Analytics
Content Analytics• Natural Language Processing• Medical Fact and Relationship
Extraction (Annotation)• Trend, Pattern, Anomaly,
Deviation Analysis
Predictive Analytics
• Predictive Scoring and Probability Analysis
Analyzed and Visualized
Information
Health Integration Framework
Data Warehouse and Model
Master Data Management
Advanced Case Management
Business AnalyticsPartners (HLI) Specialized Research
IBM Watson for Healthcare
Confirm hypotheses or seek alternative ideas with confidence based responses from learned knowledge*
* Future optional capability
IBM Content and Predictive Analytics for Healthcare
How it works
Working togetherIBM Content and Predictive Analytics for Healthcare and IBM Watson for Healthcare
48
Leverage learned knowledge with QA-style interactions for clinical applications such as diagnosis
IBM Watson for Healthcare
WellPoint and IBM Announce Agreement to Put Watson to Work in
Health CareSeptember 12, 2011
“… clinical best practices to help a physician advance a
diagnosis and guide a course of treatment"
Books, clinical guidelines, web resources, journals
and other healthcare authoritative resources
Evidence Based Learned Knowledge
Past, present and future analysis compliments Watson – with focus on customer data for clinical and operational outcomes
IBM Content and Predictive Analytics for Healthcare
ClinicalOutcomes
OperationalOutcomes
ClinicalOutcomes
OperationalOutcomes
Ready for Watson
IBM Content and Predictive Analytics for Healthcare
What’s so innovative?
A 42-year old white male presents for a physical. He recently had a right hemicolectomy invasive grade 2 (of 4) adenocarcinoma in the ilocecal valve was found and excised. At the same time he had an appendectomy. The appendix showed no diagnostic abnormality.
Patient Age: 42Gender: MaleRace: White
Procedure hemicolectomydiagnosis: invasive adenocarcinomaanatomical site: ileocecal valvegrade: 2 (of 4)
Procedure appendectomydiagnosis: normalanatomical site: appendix
* Future capability
Accurately extract buried medical facts and relationships with medical annotators
Analyze compiled information for trends, patterns, deviations, anomalies and relationships in aggregate to reveal new insights with content analytics
Model, score and predict the probability of outcomes with predictive analytics
Make insights accessible and actionable for all clinical and operational knowledge workers (and systems)
PhysiciansOther CliniciansCare CoordinatorsResearchers
ExecutivesBusiness AnalystsClaimsFraud
Other Systems and Applications
Knowledge Workers
Confirm hypotheses or seek alternative ideas from learned knowledge via Watson for Healthcare from the same user interfaces*
49
5050
Seton Healthcare FamilyReducing CHF readmission to improve care
Business ChallengeSeton Healthcare strives to reduce the occurrence of high cost Congestive Heart Failure (CHF) readmissions by proactively identifying patients likely to be readmitted on an emergent basis.
What’s Smart?IBM Content and Predictive Analytics for Healthcare solution will help to better target and understand high-risk CHF patients for care management programs by:
Smarter Business Outcomes• Seton will be able to proactively target care
management and reduce re-admission of CHF patients.
• Teaming unstructured content with predictive analytics, Seton will be able to identify patients likely for re-admission and introduce early interventions to reduce cost, mortality rates, and improved patient quality of life.
IBM solution• IBM Content and
Predictive Analytics for Healthcare
• IBM Cognos Business Intelligence
• IBM BAO solution services
• Utilizing natural language processing to extract key elements from unstructured History and Physical, Discharge Summaries, Echocardiogram Reports, and Consult Notes
• Leveraging predictive models that have demonstrated high positive predictive value against extracted elements of structured and unstructured data
• Providing an interface through which providers can intuitively navigate, interpret and take action
“IBM Content and Predictive Analytics for Healthcare uses the same type of natural language processing as IBM Watson, enabling us to leverage information in new ways not possible before. We can access an integrated view of relevant clinical and operational information to drive more informed decision making and optimize patient and operational outcomes.”
Charles J. Barnett, FACHE, President/Chief Executive Officer, Seton Healthcare Family
Featured in
The Data We Thought Would Be Useful … Wasn’t
• 113 candidate predictors from structured and unstructured data sources
• Structured data was less reliable then unstructured data – increased the reliance on unstructured data
New Unexpected Indicators Emerged … Highly Predictive Model
• 18 accurate indicators or predictors (see next slide)
Predictor Analysis % EncountersStructured Data
% Encounters Unstructured Data
Ejection Fraction (LVEF) 2% 74%
Smoking Indicator 35%(65% Accurate)
81%(95% Accurate)
Living Arrangements <1% 73%(100% Accurate)
Drug and Alcohol Abuse 16% 81%
Assisted Living 0% 13%
What Really Causes Readmissions at Seton
Key Findings
51
97% at 80th percentile
49% at 20th percentile
What Really Causes Readmissions at Seton
Top 18 Indicators
52
18. Jugular Venous Distention Indicator
17. Paid by Medicaid Indicator16. Immunity Disorder Disease Indicator15. Cardiac Rehab Admit Diagnosis with CHF Indicator14. Lack of Emotion Support Indicator13. Self COPD Moderate Limit Health History Indicator12. With Genitourinary System and Endocrine Disorders11. Heart Failure History10. High BNP Indicator9. Low Hemoglobin Indicator8. Low Sodium Level Indicator7. Assisted Living (from ICA Extract)6. High Cholesterol History5. Presence of Blood Diseases in Diagnosis History4. High Blood Pressure Health History3. Self Alcohol / Drug Use Indicator (Cerner + ICA)2. Heart Attack History1. Heart Disease History
New Insights Uncovered by Combining Content and Predictive Analytics
• LVEF and Smoking are significant indicators of CHF but not readmissions
• Assisted Living and Drug and Alcohol Abuse emerged as key predictors (only found in unstructured data)
• Many predictors are found in “History” notations and observations
Patient X was hospitalized 6 times over an 8 month period. The same basic information was available at each encounter and Patient X’s readmission prediction score never dropped below 95 (out of possible 100)
53 © 2011 IBM Corporation
The Impact of Readmissions at Seton
CHF Patient X – What Happened?Admit / Readmission
30-Day Readmission
Patient X (DSS & Cerner)High Model Score (100)Gender: MaleAge: 73Insurance: MedicaidLack of Emotional Support: YesSodium Level: LowCholesterol Level: HighCOPD History: YesHeart Disease & Heart Failure History: Yes HBP History: Yes
NLP Clinical DocumentationLiving Arrangement: Permanent Assisted Living: No Smoking History: YesSmoking Amount: N/AAlcohol Abuse History: YesDrug Abuse History: N/AEjection Fraction: N/A
Apr-18-2009 May-12-2009 May-20-2009 Oct-11-2009 Nov-24-2009 Dec-20-2009
8 days24 days 144 days 44 days 26 days
Individual Patient Data at Each Encounter (Patient X @ Dec 20, 2009)
98% 98% 96% 95% 96% 100%
Patient Population Monitoring Clinical and
Operational Data
54 © 2011 IBM Corporation
The Impact of Readmissions at Seton
CHF Patient X – What Happened?Admit / Readmission
30-Day Readmission
98% 98% 96% 95% 96% 100%
Apr-18-2009 May-12-2009 May-20-2009 Oct-11-2009 Nov-24-2009 Dec-20-2009
8 days24 days 144 days 44 days 26 days
Individual Patient Data at Each Encounter (Patient X @ Dec 20, 2009)
Patient X was readmitted the 5th time after 26 days with additional risk factors. It surfaced that there was of lack of emotional support plus Patient X had taken up smoking again as well as alcohol abuse.
Patient Population Monitoring Clinical and
Operational Data
55 © 2011 IBM Corporation
The Impact of Readmissions at Seton
CHF Patient X – What Happened?Admit / Readmission
30-Day Readmission
Patient Population Monitoring Clinical and
Operational Data
Summary of Key Readmission Risk Factors for Patient X
• Possible Intervention Factors: High Cholesterol, Low Sodium, Emotional Support, High Blood Pressure
• Other Factors: Paid by Medicaid, History: COPD, Heart Disease and Heart Failure
• A number of the top 18 factors were not available from the data at each encounter including the top predictor (Jugular Venous Distention Indicator)
Apr-18-2009 May-12-2009 May-20-2009 Oct-11-2009 Nov-24-2009 Dec-20-2009
8 days24 days 144 days 44 days 26 days
17% of Out-of-Pocket
Costs at 1st Encounter
83% of Out-of-Pocket Costs Were Avoidable (5 Unnecessary
Encounters)
Question What is Known Complement ICPA with Watson for
Healthcare to get real time, confidence based answers with evidence based learning
56
Analyze and Visualize the Past
Understand trends, patterns, deviations, anomalies,
context and more in large corpuses of historical clinical and operational information to reveal new insights
Predict the FutureUse predictive models and
scoring to make more informed decisions through
predictive and future scenario modeling
See the PresentAnalyze and extract text from in-process documents or other information to find structured
data errors … feed the results to other cases and systems
IBM Content and Predictive Analytics … Ready for WatsonComplements IBM Watson to analyze and visualize past, present and future scenarios in context
Predictive Analytics in Healthcare
57
Operational Data
ClinicalData
Access Raw Unstructured and Structured Data
ICPABeyond CHF
Information Technology
BioinformaticsResearch
Clinical Care ExecutiveBilling / Admin
• Biomedical Research• Treatment Protocols and
Effectiveness• Resource Utilization and
Operational Planning
Helps Enable:• Diagnostic Assistance• Pre-authorization
Approval
Individual or Patient Population Analysis
Applications
Create and Store Evidence Based Analyzed Information
• Trends and Patterns• Anomalies and Deviations• Patient Similarities and
Cohorts• Resource Utilization and
Cost Analysis
• Preventable Admissions• Preventable Readmissions• Early Onset Prediction• Length of Stay• Condition Deterioration
Helps Enable:• Care Coordination and
Discharge Follow-up• Claims and Analysis• Fraud Detection• Call Center / Voice-of-
Patient
Enables Better Clinical and Operational Outcomes
ClinicalOutcomes
OperationalOutcomes
ClinicalOutcomes
OperationalOutcomes
• Improve Patient Care• Competitive Advantage
and Differentiation• Optimize Resources
• Increase Revenue• Avoid Penalties• Lower Costs• Lower Risks
How to get started
58
IBM Content and Predictive Analytics for Healthcare
IBM BAO Solution Services – Center of Excellence
Also available as a Workload Optimized System
Expand solution value by integrating other systems and capabilities
Maximize solution value by extending with IBM Watson for Healthcare for real-time confidence based answers
NOW FUTURE
NOW
Address pressing clinical and operational issues today
IBM BAO Enterprise Services
Advanced Case Management
Business Analytics
Data Warehouse and Healthcare Data Model
Master Data Management
Partner Solutions
Uni
que
Val
ue
Del
iver
ed
Value Maximized
Expand and integrate ICPA-based solutions
Complement with IBM Watson for Healthcare
Start with IBM Content and Predictive Analytics1 2 3
KEY
Future Optional Capability
ICPA for Healthcare
Optionalcapabilities
ClinicalOutcomes
OperationalOutcomes
ClinicalOutcomes
OperationalOutcomes