Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.

Anurati Mathur & Propeller Health @ Madison's Big Data Meetup

1,040 views

Published on

Using healthcare data - context & considerations for collecting, cleansing, analyzing, and displaying geospatial and temporal data, with a focus on Propeller Health's program in Louisville, KY.

Published in: Healthcare
  • Be the first to comment

Anurati Mathur & Propeller Health @ Madison's Big Data Meetup

  1. 1. USING HEALTHCARE DATA Considerations, context, and storytelling August 25, 2015 Anurati Mathur @anurati
  2. 2. Propeller Health •  Founded in 2010 •  Located in Madison, WI and San Francisco, CA •  FDA 510(k) cleared •  25+ commercial programs •  Validated in 12 clinical studies totaling over 1,200 patients
  3. 3. Propeller Health
  4. 4. Cost of asthma & COPD Asthma & COPD are the 5th and 6th most expensive diseases in the US
  5. 5. Disease management has not changed in decades Daily diaries and paper-based action plans add burden & complexity and are rarely used by patients.
  6. 6. Real-time data can help detect exacerbations •  Half of patients who go to ED have had symptoms for > 4 days •  With real-time data, we can intervene early while outpatient treatment is still an option
  7. 7. What is medication non-adherence? $290B Motivations Outcomes
  8. 8. Digital interventions improve adherence Significant literature supports the use of SMS and audiovisual reminders to improve adherence to inhaled controller medications
  9. 9. How the platform works
  10. 10. Sensors measure symptoms & track symptoms Rescue Used to address symptoms when they occur unexpectedly – among the most important markers of impairment & risk Controller / maintenance Should be used regularly to minimize or eliminate symptoms by addressing underlying pathophysiology
  11. 11. Propeller passively collects data on when, where and how often a person uses their rescue inhaler
  12. 12. 1:38:08 PM PST Pollen Count 5.2 AQI 102 N47o 34.4452. W110o77.8257 2 inhalations of ProAir (120mcg) 20 seconds apart Winds from SW, 13mph And attaches meta data to each actuation
  13. 13. Multi-faceted approach to improving adherence •  Non-adherence is complex and multi-dimensional •  Improving adherence requires an equally multi-dimensional solution
  14. 14. Rules driven notifications & alerts Goal is to give users small bits of actionable, personalized feedback to foster self-management, without increasing the effort required of users •  Passive •  Feed of personalized insights •  Personalized encouragement •  Triggered questions & surveys •  Messaging from Care Team
  15. 15. 25+ commercial programs across the country
  16. 16. Retain Users Enroll Users●  Intercept during normal clinical care ●  Telephone to schedule visit ●  Email marketing ●  Direct mail ●  Online targeted marketing ●  Incentives ●  Partnership ●  In person (face-to-face) ●  On telephone ●  Online self-enrollment ●  Hardware ●  Software ●  CSS Team Experiments every step of the way
  17. 17. The healthcare system Patient
  18. 18. The healthcare system Patient MD, NP, RN Hospital, clinic Provider $$
  19. 19. The healthcare system Patient MD, NP, RN Hospital, clinic Provider $$ 1. Goods 2. Services Time Expertise
  20. 20. The healthcare system Insurance companies MD, NP, RN Hospital, clinic Provider 1. Goods 2. Services Time Expertise $$ Patient
  21. 21. The healthcare system Insurance companies MD, NP, RN Hospital, clinic Provider 1. Goods 2. Services Time Expertise $$ $$$$$$ $$ $$ Patient
  22. 22. The healthcare system Insurance companies MD, NP, RN Hospital, clinic Provider 1. Goods 2. Services Time Expertise Deductible, copay $$$$$$ $$ $$ $$$$ Risk-sharing Patient
  23. 23. The healthcare system Insurance companies MD, NP, RN Hospital, clinic Provider 1. Goods 2. Services Time Expertise Deductible, copay $$ $$$$ Patient Worker $$ $$
  24. 24. The healthcare system Insurance companies MD, NP, RN Hospital, clinic Provider 1. Goods 2. Services Time Expertise Deductible, copay $$ $$$$ Patient Worker Citizen $$ $$ $$
  25. 25. The healthcare system Insurance companies MD, NP, RN Hospital, clinic Provider 1. Goods 2. Services Time Expertise Deductible, copay $$ $$$$ Patient Worker Citizen $$ $$ $$ $$ $$ $$
  26. 26. The healthcare system Insurance companies MD, NP, RN Hospital, clinic Provider 1. Goods 2. Services Time Expertise Deductible, copay $$ $$$$ Patient Worker Citizen $$ $$ $$ $$ $$$$
  27. 27. Healthcare goods & services Pharma, digital health, and more
  28. 28. The healthcare system Insurance companies MD, NP, RN Hospital, clinic Provider 1. Goods 2. Services Time Expertise Deductible, copay $$ $$$$ Patient Worker Citizen $$ $$ $$ $$ $$$$
  29. 29. Healthcare data – until recently Insurance companies MD, NP, RN Hospital, clinic Provider 1. Goods 2. Services Time Expertise Deductible, copay $$ $$$$ Patient Worker Citizen $$ $$ $$ $$ $$$$ 1 2
  30. 30. Data are the backbone of value-based care Better data are needed to understand what happens to patients in between physician visits •  to effectively manage care •  to improve outcomes •  to prove effectiveness
  31. 31. Healthcare data - today Insurance companies MD, NP, RN Hospital, clinic Provider 1. Goods 2. Services Time Expertise Deductible, copay $$ $$$$ Patient Worker Citizen $$ $$ $$ $$ $$$$ 1 2 4 3
  32. 32. Healthcare data - today
  33. 33. Healthcare data - today
  34. 34. Healthcare data - today
  35. 35. Other, relevant non-healthcare data
  36. 36. •  Enacted August 21, 1996 •  4 key components, including: administrative simplification •  Administrative simplification, including: privacy & security standards Health Insurance Portability and Accountability Act Privacy standards •  Protected Health Information (PHI) •  Covered entities •  Routine purposes Security standards •  Administrative: Policies, training, etc. •  Physical: unauthorized intrusion, etc. •  Technical: encryption, etc.
  37. 37. Once information is de-identified, it is no longer subject to the privacy regulation restrictions. De-identification Statistical de-identification •  Statistical expert confirms that data could not reasonably be re-identified •  Documentation & results of all statistical tests & analyses Safe Harbor Guidelines Removal of 18 personal identifiers that constitute Personally identifiable information (PII). 1.  Names 2.  Geo-specificity < than state 3.  All dates (birth, death, doc visits, etc.) 4.  Telephone numbers 5.  Fax numbers 6.  Email addresses 7.  Social security #s 8.  Medical record #s 9.  Health plan #s 10.  Account #s 11.  License #s 12.  Vehicle identifiers 13.  Device identifiers 14.  URLs 15.  IP addresses 16.  Biometric identifiers 17.  Full face photography 18.  Any other uniquely identifying #
  38. 38. Statistical de-identification
  39. 39. Statistical de-identification
  40. 40. Clinical trials Series of statistically instrumented experiments to prove safety & efficacy of treatments, devices, and medical strategies Development process – medical devices
  41. 41. Propeller clinical trials design 500-1,000 enrolled Intervention Control 6-12 month study period, with quarterly assessments of control Randomization Sensor + intervention Sensor + no intervention Primary outcomes Asthma control Poorly controlled Not well controlled Well controlled Inhaler use Rescue inhaler Controller inhaler
  42. 42. Louisville clinical results Rescue inhaler use over time Asthma control over time
  43. 43. Case study: Propeller Health in Louisville, Kentucky “One of the top 20 “most challenging places to live with asthma in the US.” - Asthma & Allergy Foundation of America, 2014
  44. 44. Snapshot of Louisville health •  1 in 3 adults in Louisville are overweight or obese -  Increased risk for high blood pressure, diabetes, some cancers, arthritis / joint pain •  25% of Louisville adults still use tobacco products -  Increased risk for lung disease, cancer, heart attack and stroke, and poor wound healing •  Louisville’s rate of deaths for heart disease, cancer, diabetes, stroke are higher than the death rates for Kentucky and the US
  45. 45. Why do this study? Asthma hospitalizations in Jefferson County, 2011 Data like this has many problems: •  Based on healthcare utilization •  Retrospective, time lags •  Aggregated, low spatial resolution
  46. 46. Louisville study design Enrollment begins Baseline data collection for 1 month instrumentation with sensors, but no access to data Intervention period & data collection for 12 months 5,430 unique rescue inhaler events recorded 27,197 person-days of data
  47. 47. Temporal data
  48. 48. Temporal data Air quality index (Levels of health concern) Numerical value Meaning Good 0 to 50 Air quality is considered satisfactory & poses little or no risk Moderate 51 to 100 Air quality is acceptable, but for some pollutants there may be risk for unusually sensitive groups Unhealthy for sensitive groups 101 to 150 Members of sensitive groups may experience health effects Unhealthy 151 to 200 Everyone may begin to experience health effects Very unhealthy 201 to 300 Health warnings of emergency conditions Hazardous 301 to 500 Health alerts: everyone may experience serious health effects
  49. 49. Temporal data
  50. 50. EPA monitoring in Jefferson County QGIS
  51. 51. Understanding the temporal data Tableau Inhaler events / person / day
  52. 52. Understanding the temporal data Inhaler events / person / day
  53. 53. Understanding the temporal data
  54. 54. Understanding the temporal data Tableau
  55. 55. Understanding the temporal data
  56. 56. Understanding the temporal data Tableau
  57. 57. Geospatial data •  Usually latitude, longitude •  A shapefile is an vector data storage format for storing the location, shape, and attributes of geographic features •  Projections matter! This slide has an animation.
  58. 58. Projecting a head
  59. 59. North America This slide has an animation.
  60. 60. World Geodetic System, 1984 – WGS 84 •  Standard for use in cartography and navigation, including GPS •  Coordinate origin at the Earth’s center of mass •  Meridian of zero longitude is ~336 feet east of Greenwich •  Originally developed by the Department of Defense in 1966 •  Default projection for Tableau & QGIS
  61. 61. Inhaler use events Tableau
  62. 62. Inhaler use events – wide distribution
  63. 63. Inhaler use events – by device type Tableau
  64. 64. Inhaler use events – by medication Tableau
  65. 65. Inhaler use events & census geography QGIS
  66. 66. Inhaler use events & census geography QGIS
  67. 67. Inhaler use events & other interesting shapefiles Total Crime Index, ESRI crime indices data, 2014 QGIS
  68. 68. Inhaler use events & other interesting shapefiles QGIS Median household income, American Community Survey (ACS) data, 2012
  69. 69. Inhaler use events & other interesting shapefiles QGIS Median year housing structure built, American Community Survey (ACS) data, 2012
  70. 70. Inhaler use events & other interesting shapefiles Metro parks, Louisville / Jefferson County Information Consortium (LOJIC), 2010     QGIS
  71. 71. Inhaler use events & other interesting shapefiles Tree canopy, National Land Cover Database / USGS, 2001 QGIS    
  72. 72. Inhaler use events & other interesting shapefiles Major roads, Kentucky Transportation Cabinet, 2015     QGIS
  73. 73. QGIS Inhaler use events & other interesting shapefiles Avg. annual traffic count, Kentucky Transportation Cabinet, 2012    
  74. 74. QGIS Inhaler use events & other interesting shapefiles Avg. annual traffic count, Kentucky Transportation Cabinet, 2012    
  75. 75. Inhaler use events & other interesting shapefiles Active mines, Kentucky Mine Mapping Information System, 2008 QGIS
  76. 76. Inhaler use events & other interesting shapefiles This slide has an animation.
  77. 77. So what? This slide has an animation.
  78. 78. Spatial predictors we explored •  Exposures •  Proximity to emission sources, industrial areas, power plants, active mines •  Transportation infrastructure •  High density traffic areas •  Land use •  Impervious surface (%) •  “Greenness” and tree canopy •  Proximity to green spaces & parks •  Neighborhood •  Socioeconomic measures: household income, educational attainment, property values, property age, crime rates •  Population density •  Study factors •  # of users enrolled per zip
  79. 79. Kernels of inhaler use density to find associations ArcGIS, ERDAS
  80. 80. Ecological niche modeling 1. Geo-reference inhaler use points 2. GIS of environmental variables 3. Extract environmental data at each point 4. Build model against background sample of pseudo-absences 5. Model calibration & selection (AUC, cross- validation with test data, etc.) 6. Probability distribution based on environmental suitability 7. Assessment of environmental variable contributions 8. Extrapolate into new environmental or geographic space 9. Test with additional data
  81. 81. Probability of inhaler use – environmental modeling ArcGIS, ERDAS
  82. 82. Target areas for intervention ArcGIS, ERDAS
  83. 83. Testing Louisville strategic intervention scenarios •  What if Louisville… •  Reduced traffic density?   •  Reduced particulate matter?   •  Greened downtown?    
  84. 84. What if Louisville achieved a 10% reduction in PM2.5 ArcGIS, ERDAS
  85. 85. à 27% reduction in amount of area at risk ArcGIS, ERDAS
  86. 86. Current AIR Louisville emphases •  Improved data & sampling •  More analyses, guided closely by city needs •  More transparency & public awareness with open data portal
  87. 87. Current AIR Louisville emphases
  88. 88. Obama Climate Data Initiative
  89. 89. Sharing Louisville data publicly Interesting / informative Surprising / delightful
  90. 90. D3 This slide has an animation.
  91. 91. D3
  92. 92. D3
  93. 93. D3
  94. 94. D3
  95. 95. D3
  96. 96. D3
  97. 97. D3
  98. 98. Visualizing inhaler events – version 1 D3, TopoJSON This slide has an animation.
  99. 99. Mapbox
  100. 100. Mapbox
  101. 101. Mapbox
  102. 102. Visualizing inhaler events – version 2 D3, Mapbox This slide has an animation.
  103. 103. Visualizing inhaler events – version 3 D3, Mapbox This slide has an animation.
  104. 104. Louisville is one of the top 20 most difficult places to live with asthma in the US. AIR Louisville enrolled 300 residents to track their asthma rescue inhaler use with Propeller Health sensors. AIR Louisville collected data on citizens’ inhaler use for 12 months Here’s what we learned… We learned a lot about how Louisville residents experience asthma symptoms and use their inhalers. Only 31% of asthma attacks occur in homes. People use inhalers more when it is hot outside. Weed, pollen and mold significantly increase asthma risk. Wind clears the air and reduces asthma attacks. When pollution levels are high, people have more asthma symptoms. Strong city partnerships are key to a successful program. People use rescue inhalers more on weekdays than on weekends. Trees provide a protective effect for asthma attacks (especially evergreens). A 10% reduction in particulate matter could mean a 27% reduction in asthma risk. Blue dot = inhaler signal This slide has an animation.
  105. 105. In Mayor Fischer’s own words This slide has an animation. https://youtu.be/iiaTPV09vBE?t=6m27s
  106. 106. Thank you! Anurati Mathur anuratim@gmail.com @anurati https://www.linkedin.com/in/anurati

×