D10/E10 Dawning of a New Epoch in Harm Measurement From Paleolithic Hunter Gathers to Holocyne Farmers Jack Jordan Henry Ford Health System (This presenter has nothing to disclose) December 10, 2014 Objectives Implement strategies to leverage new EMRs to make action on harm visible and actionable within 48 hours or less Share a cutting-edge method for comprehensive, real-time harm measurement Engage you in a journey to re-invent harm measurement 1
Analytics in Baseball Baseball has always been a sport of numbers People have been keeping statistics on baseball for over 100 years Until recently, big data on baseball was unavailable Now every pitch is tracked and analyzed This has changed how the game is played Healthcare Analytics is going thru this same paradigm Babe Ruth s Information 2
David Ortiz s Information 3
MLB Batting Averages Over Time Analytics Making an Impact Back to the Real World Measurement of Harm in hospitals is low resolution, delayed and rarely actionable. We have spent $100s of Billions on new EMR What good ideas do we have for leveraging that with analytics? 4
Goal of HARM 2.0 Comprehensive tracking of harm by Oct 2015 with data within 48 hours of triggering documentation with no human intervention. Harm may include iatrogenic vulnerability as well as harm requiring additional treatment Real Goal of Program Give tools for insight and action to front line staff and middle management. Make gaps in care visible and actionable Make not testing changes seem very uncomfortable. 5
What is Comprehensive? Medication DVT Acute Renal Failure Blue Alert Pneumothorax Puncture/Laceration Unexpected blood use post Procedure Aspiration Pneumonia Other Procedural complications Infections Pressure Ulcer Falls Patient Trauma Other Procedural Complications Environment Hypoglycemia Anticoagulation issues (INR > 5) Narcan Diuretics causing adverse effects Allergic Reaction not POA C-diff toxin positive Delirium GI Bleed not POA SSI CAUTI CLABSI Pneumonia VAE Other Pneumonia Not POA Perinatal Ideal Delivery Meaningful Use and Available Data Traditional EMR Era ICD9 Dx ICD9 PX Cpt4 (maybe) Limited Labs/cultures LOS Charges ADT locations Individual Charge master items All Traditional Problem lists (maybe) Orders Medication Administration Vital Signs (limited) Flow Sheet data Equipment feeds (maybe) 6
What is Special about HFHS Problem based charting Long History of Quality Improvement Open Data Environment Data Reporting & Analytics are not part of IT New EMR with all hospitals on the same build Problems in Paradise Definitions are far more complicated Audiences are different for data with new distribution channels Choices Comorbid Edema (from Flow Sheet?, problem list? Past ICD9 code, Medications?) 7
What Have We Learned So Far? Timely delivery changes the intervention from the ground up Related opportunities appear in the process Some traditional measures not useful Predictive Analytics are not as valuable as actionable analytics Weakness in the data usually uncover other interesting opportunities in Patient Care What We are Learning Measurement becomes tightly coupled to the care-giving Design of the documentation is profoundly linked to data possibilities 8
Delivery and Follow up Traditional Monthly reports to leadership EMR Advanced Detail to front-line Detail lists for deep dives Roll up with analysis to leadership Teams built around project and data from team out to staff Detail can be both for team and front line real time Detail from the Beginning 9
Example VTE Harm: Problem list (Added during stay) low resolution, non-deep Vein, at risk for vs real, POA reliability Treatment received (Heparin drip, etc) Logic to weed out a-fib, etc. used for Drip Imaging results (CT, Venous Doppler and duplex) Order but No results available in Clarity Heart problems (Problem list ICD9 codes 410.xx and 427.xx) Billing data (ICD9 code) Not used in logic Lab results N/A VTE is Really Complicated No single variable is good enough Treatment overlaps with other problems Numerous patients with Heparin or Lovenox and no legitimate problem on problem list 10
Real-Time Method: Review Identification within 48 hours of documentation Real time identification through the artifacts of care Problems list Ordering a treatment/medication Lab value Accuracy and reliability Insight into the variation of practice Organic system allows faster response to change Special case problems Switch from Warfarin to Heparin for existing VTE at admission If VTE is added to problem list after admission may cause false positive Other reason for treatment and VTE on problem list One case with A-fib and treatment still ended up with DVT Start treatment and stop after study (reversal) Continuing to treat a superficial vein thrombosis after study 11
DVT/PE Harm: Logic VTE: Chart Reviews 427 Chart Reviews were done Comparison: Proposed Real Time Logic: Sensitivity: 84.4% Specificity: 97.0% Modified PSI#12 Sensitivity: 37.5% Specificity: 99.0% Positive Predictive Value Negative PredictiveValue 69%Likelihood ratio positive 99%Likelihood ratio negative 27.98 True Positive 0.16 False Negative 27 False Positive 5 True Negative 12 386 Positive Predictive Value Negative PredictiveValue 75%Likelihood ratio positive 95%Likelihood ratio negative 37.31 True Positive 0.63 False Negative 12 False Positive 20 True Negative 4 394 We are finding a little bit more, but wait! Chart Reviews Proposed Logic Billing Total Dif. 32 0% 39 22% 16 50% Improvement in documentation can significantly improve accuracy 12
VTE Harm: Actual vs. Modified PSI#12 vs. Logic 2 2 10 10 2 17 3 Actual Chart Reviews False Positive? 81 y.o. male c/o sudden worsening right sided chest pain. Pt was recently discharged local hospital for acute on chronic respiratory failure due to HCAP and mucus plugging. Pt. with hx of PE 25 years ago and no longer on Coumadin VQ scan negative but PE remains on problem list Pt with hx DVT and now he has malignancy and getting chemo. He has pleuritic chest pain and he was hypoxic initially. Pt will need to be on long term lovenox Doppler negative for DVT but remains on problem list 13
Alternative Logic: Receiving Meds and No Heart Problem Modified PSI#12 Sensitivity: 37.5% Positive Predictive Value 75%Likelihood ratio positive Specificity: 99.0% Negative PredictiveValue 95%Likelihood ratio negative Alternative Logic: Receiving Meds and No Heart Problem Sensitivity: 71.9% Positive Predictive Value 15%Likelihood ratio positive Specificity: 67.3% Negative PredictiveValue 97%Likelihood ratio negative 37.31 True Positive 0.63 False Negative 12 False Positive 20 True Negative 4 394 2.20 True Positive 0.42 False Negative 23 False Positive 9 True Negative 130 268 Using: Problem List Medication Administration Records Pros & Cons: Better sensitivity but worse PPV Worse specificity but better NPV Lots of false positives Alternative Logic: Only Problem List Modified PSI#12 Sensitivity: 37.5% Positive Predictive Value Specificity: 99.0% Negative PredictiveValue Alternative Logic: Only Problem List Sensitivity: 93.8% Positive Predictive Value Specificity: 87.7% Negative PredictiveValue 75%Likelihood ratio positive 95%Likelihood ratio negative 37.31 True Positive 0.63 False Negative 12 False Positive 20 True Negative 4 394 38%Likelihood ratio positive 99%Likelihood ratio negative 7.61 True Positive 0.07 False Negative 30 False Positive 2 True Negative 49 349 Using: Only Problem List Pros & Cons: Better sensitivity but worse PPV Worse specificity but better NPV Almost picking all the cases but significant number of false positives Can be used in case only Problem list is available 14
Alternative Logic: Ignoring Problem List Modified PSI#12 Sensitivity: 37.5% Positive Predictive Value Specificity: 99.0% Negative PredictiveValue Alternative Logic: Ignoring Problem List Sensitivity: 87.5% Positive Predictive Value Specificity: 75.9% Negative PredictiveValue 75%Likelihood ratio positive 95%Likelihood ratio negative 37.31 True Positive 0.63 False Negative 12 False Positive 20 True Negative 4 394 23%Likelihood ratio positive 99%Likelihood ratio negative 3.63 True Positive 0.16 False Negative 28 False Positive 4 True Negative 96 302 Using: Only using Medication Administration Records Pros & Cons: Better sensitivity but worse PPV Worse specificity but better NPV Only 4 false negatives but significant number of false positives Can be used in case problem list is not available Alternative Logic: Ignoring POA condition (Meds in first 24hrs) Modified PSI#12 Sensitivity: 37.5% Positive Predictive Value 75%Likelihood ratio positive 37.31 True Positive Specificity: 99.0% Negative PredictiveValue 95%Likelihood ratio negative 0.63 False Negative Alternative Logic: Ignoring POA condition ( Receiving Treatment in the First 24hrs of Admission) Sensitivity: 90.6% Positive Predictive Value 49%Likelihood ratio positive 12.02 True Positive Specificity: 92.5% Negative PredictiveValue 99%Likelihood ratio negative 0.10 False Negative 12 False Positive 20 True Negative 4 394 29 False Positive 3 True Negative 30 368 Using: Problem List Medication Administration Records Pros & Cons: Better sensitivity but worse PPV Worse specificity but better NPV Only 3 false negatives but more false positives for pre-existing DVT/PE cases 15
VTE Harm/1000 patient days: HFHS Total VTE Harm/1000 Patient Days: Estimated Real Harm: HFHS Total 3.0 2.5 2.0 1.5 Logic: HFHS Total Billing: HFHS Total 1.0 Estimated Real Harm 0.5 0.0 Other Lessons and Data Failures in the measurement of DVT are tightly connected to practice issues Building reports on use of Doppler & CT scans per found DVT Continued treatment of superficial vein clots needs feedback loop Timelines don t match Date of discharge vs Date of problem in hospital 16
How will we use these data? EPIC Radar Dashboards do not allow for definitions this complicated Developing dashboards and dedicated Data marts Work with build team to leverage lessons Channels for the data are far more complicated Nursing data Care team Project Team GME data for education Pressure Ulcers Current tracking based on monthly prevalence audit Numerous real time needs Current Patient List at the touch of a button Stage 3 and above? Not Present on admission Braden score below 18 Gaps in documentation List of all patients for study 17
New Opportunities How often do Ulcers Progress? Real time reliability of POA documentation Gaps in documentation found Size can be evaluated when documented. Variation in documenting size Inches, cm, quarter sized, etc. What is the status of a Pressure Ulcer at Discharge? (do you keep it in the record?) Areas for Improvement Can we improve the reliability of ordering Pressure Ulcer Prevention (PUP) protocol NDNQI (Hospital Acquired vs Unit Acquired) This causes problems in documentation 18
Pressure Ulcer EMR Detailed Pressure Ulcer Harm: Audits vs. Real-time (Cases per month) 250 200 Audits (est.) 150 Real-Time: including POA Missing but > 24hrs 100 Real-Time: NOT POA 50 0 May 14 June 14 July 14 Aug 14 Sep 14 Oct 14 19
Hospitals have Different Drivers 30% 25% 20% 15% 10% 5% 0% HFHN HENRY FORD HOSPITAL HFMH MACOMB HOSPITAL HFWB WEST BLOOMFIELD HOSPITAL HFWH WYANDOTTE HOSPITAL (Percent of Pressure Ulcers on a Patient that received a Vasopressor) New Questions Can Be Answered Incontinence Associated Dermatitis vs. Stage II Healing stage can be assessed Predictive analytics on progression planned 20
Some Changes Require Altering Build of EMR Using Best Practice Alert (BPA) to initiate protocol for patients with Braden < 18 Making Incontinence Associated Dermatitis (IAD) easier to document accurately Improving documentation of size Model for Skin Break Down Database on wound nurse computer Reports to reduce labor for NDNQI audits List of patients in house with ulcers Reports to compare with audits and to automate audit supporting data 21
Failure to Rescue All Transfers from MED/SURG or observation to ICU Last vitals (BP, Pulse, RR), hours from Admission, max lactate prior to transfer Was RRT called? Lessons in ICU Transfer First 24 hours different than after 24 hours Average of LACT 2.5+Column Labels Row Labels >24 HFH 29.1% HFMH 14.1% HFWH 6.9% WBH 13.6% Grand Total 23.8% <24 Grand Total 14.8% 25.9% 17.1% 14.8% 7.5% 7.1% 15.4% 14.1% 14.3% 21.4% 22
Growing into a Complicated Solution Each measure seems to grow a parallel set of process data tracking Each solution is different All this data may become overwhelming Challenges We Already Know Many of the next topics require reframing for maximum benefit. Multiple vectors of harm require separate data but are intertwined. Anticoagulants and GI bleed 23
Delirium Finding patients with Delirium may be counter productive Tracking opportunities may be better strategy Patients over 80 or with Dementia receiving Benadryl or Benzodiazepines or Ambien Paying attention to eye glasses and hearing Infections NHSN has defined current measures but real time data may cause reframing Hospital acquired Pneumonia vs. VAE UTI may have poor resolution and take a couple iterations False positives may have more benefit in antimicrobial stewardship than infection reduction 24
Blood and Bleeding Teasing out unexpected drop in Hgb or blood use Linking bleeding with anticoagulants (INR>5) Attempting to integrate tracking of bleeding with good management of blood products. Medication Glucose, Anticoagulants, and Opioids All Others are difficult Anaphylaxis Hives \ Allergic reaction Rare events 25
Local Next Steps Desire to replace current No Harm measure with advance measure rather than build with ICD10 Figure out distribution channels to support each issue Procedural Harm / Infection control Grand Next Steps Find kindred spirits doing similar work to learn together Share tools and methods across the country Impact thinking of policy makers to rethink how e-measures are developed. 26
Current State NQF NQF process is NOT designed for this model of measurement Assumption that measures must be common Slow to adopt e-measures Main purpose of measures to rate and score delivery of care Severity adjustment very difficult at national level Source of data billing or submission What Does all This Mean? Common data could come from information exchanges with some loss of resolution Common performance on fixed scenarios possible instead of common measures Increasing detail makes the divergence between measures to improve and measures to rate bigger Sources of data are orders of magnitude larger and work for parsimonious source will take time 27
Questions, Thoughts What excites you about this work? Contact Info Jack Jordan jjordan1@hfhs.org 313-874-4988 S. Mani Marashi smarash1@hfhs.org 313-808-8424 Linkedin 28