Socio-behavioral determinants of health (SBDH) – a combination of behavioral, social, economic, environmental, and occupational factors – are powerful drivers of morbidity, mortality, and future well-being of individuals and communities, yet they mostly lie outside domain of the conventional medical care delivery system. Modifiable behaviors and exposures that occur in the community play a significant role in 60% of preventable deaths in the U.S.¹

Social and Non Clinical Risk Factors²

Project Executive Summary

In 2017, the Institute for Clinical and Translational Research’s (ICTR) Behavioral, Social and Systems Science (BSSS) Translational Research Community (TRC) advisory board funded this project to examine the availability of social and behavioral data in JHMI’s EPIC EMR/PHR systems. Both researchers and administrators recognize that patients’ socio-behavioral determinants play a critical role their care experiences and outcomes. Being located in Maryland with its global budgets and population-based reimbursement scheme, it is advantageous for JHU/JHMI to find cost-effective, community-level solutions that improve the population’s health status. The vision of the BSSS TRC board is to enable JHU researchers to utilize social and behavioral data collected from JHMI patients and stored in various data sources such as EPIC.

EHR data (i.e., Epic) will play an instrumental role in population health management efforts of value-based providers such as JHMI^[3-9]. EHR-extracted socio-behavioral determinants data can potentially help to coordinate care and risk stratification efforts ^[10]. However, certain challenges will remain while extracting socio-behavior factors from JHMI’s Epic such as: data quality issues and missing socio-behavioral data ^[11-12], ongoing immaturity of EHR’s advanced functionality across various providers ^[13], and the need for complex methods to extract socio-behavioral determinants from EHR’s unstructured text ^[14].

Given the increasing alignment of population and public health efforts ^[15-18], identifying socio-behavioral factors of high-risk JHMI patients will be key in addressing underlying disparities within JHMI’s population as Maryland’s all-payer waiver program is already piloting population-level outcome measures ^[19-20]. Using non-Epic data sources, such as CRISP information (i.e., Maryland’s health information exchange), can also be used to extract socio-behavioral determinant data ^[21].

For the purposes of this project, Epic is the sole data source. No legacy or ancillary systems were queried. As of 7/1/2018, there are approximately 5.4 million unique patients in Epic. The contents of this page will be reviewed and updated regularly.

Phase I Social Project Guide

Phase II Social Project Guide

Data Availability Timeline

The following diagram depicts the approximate timeline for when data was migrated into Epic or when Epic began capturing data, segmented by data type and facility.

Data Variables

• Address/Zip Code
• Alcohol Use
• Ethnicity
• Income/Financial Issues
• Housing Issues
• Language
• Race
• Smoking Status
• Social Support

Address/Zip Code

Most Common Collection Method

The most common method to collect patient permanent address is upon registration of each encounter. Address is defined as a street address/number, an optional line for apartment or other information, a city, a state or province, and a zip code.

General Completeness Rate

95% – out of approximately 5.4 million patients, 5.2 million patients have indicated a permanent address using this collection method.

Collection Date Range

Permanent address was migrated into Epic beginning in 2003 and is currently collected.

Facility Type (Inpatient or Outpatient)

Permanent address is collected at all facilities at the time of registration.

Address History

In 66% of the 5.4 million patients in Epic, there are address change records available. Effective start and end dates are included to track address changes over time. Address change records were first collected in 2003 and are currently tracked.

Other Collection Methods

In addition to the most commonly collected method, patient address is captured in the following cases:

Alcohol Use

Most Common Collection Method

Alcohol use is captured in the Social History portion of Epic during a patient encounter, whether in-person or non in-person encounters (telephone, MyChart, documentation). For the purposes of this data catalog, the variable of focus is “Alcohol Drinks Per Week”.

Collection Date Range

Social History has been collected since April 2013 and is currently collected.

General Completeness Rate

Of the approximately 5.4 million patients, 490,348 (.09%) patients reported having any value (including 0 alcoholic drinks per week) in social history. 178,789 (.03%) reported having one or more alcoholic drinks per week.

Facility Type (Inpatient or Outpatient)

Alcohol use is collected at all facilities at the time of an encounter.

Other Collection Methods

For future revisions of this data catalog, alcohol use data collected from flowsheets, questionnaires, and clinical notes will be included.

Ethnicity

Epic captures a patient’s ethnicity separate from a patient’s race, and ethnicity is defined as either Hispanic or Not Hispanic.

Most Common Collection Method

The most common method to collect patient ethnicity is upon registration of each encounter.

General Completeness Rate

50% – out of approximately 5.4 million patients, 2.7 million patients have indicated ethnicity using this collection method.

Collection Date Range

Ethnicity has been collected in Epic since 2003 and is currently collected.

Facility Type (Inpatient or Outpatient)

Ethnicity was migrated into Epic beginning in 2003 and is currently collected.

Other Collection Methods

In addition to the most commonly collected method, ethnicity is captured in the following cases:

Income/Financial Issues

Patients with income/financial issues are patients in deteriorated financial status, financial hardship, or in poverty. They are unable to afford the basics of life and/or medical interventions. They are in need and eligible for any benefit or enrolment in financial assistant programs.

Collection Methods

In order to determine the population and distribution of patients having income/financial issues, several collection methods were utilized.

Diagnosis Codes (ICD-10)*

Searching Epic for patients having ICD-10 coded diagnoses on the problem list, billing codes, or recorded at the time of an encounter yielded the following results:

* Patients with ICD-9 coded diagnoses are also included in this query. The timeframe is January 1, 2003 through June 26, 2018.

Clinical Notes

Using Natural Language Processing (NLP) techniques, keyword phrases indicating specific SDH were extracted from 23 million clinic notes authored by specific provider types between July 1, 2016 and May 31, 2018. The notes represent 1,188,202 unique patients and 9,066,508 unique encounters.³

We pre-processed the notes to mark the line boundaries and remove non-Unicode characters. To identify notes containing SDH, we used hand-crafted linguistic patterns developed by a team of experts based on SDH descriptions in LOINC, SNOMED, ICD10, Public Health Surveys and Instruments (ACS, American Housing Survey, NHANE, etc.), phrases from the literature review and other studies, and manual tagging.

To craft the linguistic patterns, the expert team focused on three domains: Housing, Finance, and Social Support.

Findings for Financial Issues:

• Of the 1,188,202 unique patients, 1% have at least one note containing mentions of financial issues.
• Of the 20,219 notes containing mentions of financial issues, about 52% were authored by physicians, 37% were authored by social workers, 8% were authored by nurses, 3% were authored by case managers and case coordinators.
• The top note types for mentions of social support are progress notes (46%), H&P notes (8%), treatment plans (6%), consults (6%), with the remaining note types dispersed among committee review, patient instructions, discharge summaries, and plan of care notes.

Housing Issues

Housing issues are categorized to those related to housing instability or insecurity, homelessness, and characteristics of the house and a number of sub-categories as follows:

• Housing Instability
  1. Having a House with Problems
  2. Prone to Homelessness

• Homelessness
  1. Currently Homeless
  2. Homelessness Being Addressed

• Housing Characteristics
  1. Quality of Building
  2. Characteristics of Building
  3. Building Environmental Health Hazards
     1. Air Quality (Including Mold)
     2. Infestation
     3. Old Paint
     4. Hazardous Material and Fire Protection
     1. Problems with House Amenities
        1. Water
        2. Sewage and Disposal
        3. Fuel and Heating/ Cooling System
        4. Electricity and Internet
        1. Age of Building
        2. Quality Problems Being Addressed

Collection Methods

In order to determine the population and distribution of patients having a housing insecurity situation, several collection methods were utilized.

Diagnosis Codes (ICD-10)*

Searching Epic for patients having ICD-10 coded diagnoses on the problem list, billing codes, or recorded at the time of an encounter yielded the following results:

* Patients with ICD-9 coded diagnoses are also included in this query. The timeframe is January 1, 2003 through May 31, 2018.

Clinical Notes

Using Natural Language Processing (NLP) techniques, keyword phrases indicating specific SDH were extracted from 23 million clinic notes authored by specific provider types between July 1, 2016 and May 31, 2018. The notes represent 1,188,202 unique patients and 9,066,508 unique encounters.³

We pre-processed the notes to mark the line boundaries and remove non-Unicode characters. To identify notes containing SDH, we used hand-crafted linguistic patterns developed by a team of experts based on SDH descriptions in LOINC, SNOMED, ICD10, Public Health Surveys and Instruments (ACS, American Housing Survey, NHANE, etc.), phrases from the literature review and other studies, and manual tagging.

To craft the linguistic patterns, the expert team focused on three domains: Housing, Finance, and Social Support. We delved into the housing domain by created patterns for its subdomains, including homelessness, prone to homelessness, homelessness being addressed, housing instability, housing characteristics – characteristics of the building, housing characteristics – the quality of the building.

Findings:

• Of the 1,188,202 unique patients, 3% have at least one note containing mentions of housing issues.
• Of the 108,439 notes containing mentions of housing issues, about 53% were authored by physicians, 32% were authored by social workers, 10% were authored by nurses, 5% were authored by case managers.
• The top note types for mentions of housing issues are progress notes (46%), ED provider notes (13%), H&P notes (8%), with the remaining note types dispersed among discharge summaries, consults, and plan of care notes.
• Manual annotation of 100 randomly selected notes containing the phrase “homeless” revealed:
  • Out of 100 notes, there are 130 mentions of the word “homeless”.
  • In 2 out of 100 notes, there were conflicting true positives and false positives within the same note.
  • In 20 out of 100 notes, there were true negatives (the note was derived from a SmartPhrase), and the answer to Homeless Y/N was No.
  • From the remaining notes, 64% were true positives, and 14% were false positives. The other notes were true negatives or unclear.

• The following numbers show the count of notes containing specific variables from the housing subdomains:
  • Homelessness Being Addressed: 82
  • Housing Instability: 5,653
  • Housing Characteristics – Characteristics of the Building:148
  • Housing Characteristics – Quality of the Building: 2,011

Questionnaires

Several questionnaires exist in Epic that capture data related to housing issues. Each is described briefly below, including frequency, general completeness rate, and facility type where the questionnaire answers were completed. Please note that data provided on general completeness reflects questions that were available to an Epic user and were answered.

Language

Most Common Collection Method

Patient preferred language is captured in Epic at the time of admission.

Collection Date Range

Preferred language was migrated into Epic beginning in 2003 and is currently collected.

General Completeness Rate

Of the approximately 5.4 million patients, 2,718,416 patients (49%) have indicated a preferred language in Epic. Of those indicating a preferred language, 2,624,122 indicated English as their preferred language.

The top 7 preferred languages, by unique patient count are as follows:

1. No language reported (null) – 2,804,973 (51%)
2. English – 2,626,379 (48.5%)
3. Spanish – 53,446 (1%)
4. Arabic – 7,317 (.1%)
5. Unknown (a valid value in Epic, different from an empty record) – 5,936  (.1%)
6. Chinese (Mandarin) – 4,036  (.1%)
7. Korean – 3,168  (.1%)

Facility Type (Inpatient or Outpatient)

Preferred language is collected at all facilities at the time of an encounter.

Other Collection Methods

For future revisions of this data catalog, preferred language collected from flowsheets, questionnaires, and clinical notes will be included.

Race

Most Common Collection Method

The most common method to collect patient race is upon registration of each encounter. Patients can select multiple races to self-identify:

• • • • American Indian or Alaska Native
      • Asian
      • Black or African American
      • Declined to Answer
      • Native Hawaiian or Other Pacific Islander
      • Other
      • Unknown
      • White or Caucasian

General Completeness Rate

90% – out of approximately 5.4 million patients, 4.9 million patients have indicated at least one race using this collection method.

Collection Date Range

Race was migrated into Epic beginning in 2003 and is currently collected.

Facility Type (Inpatient or Outpatient)

Race is collected at all facilities at the time of registration.

Other Collection Methods

In addition to the most commonly collected method, race is captured in the following cases:

Smoking Status

Most Common Collection Method

Smoking status is captured in the Social History portion of Epic during a patient encounter, whether in-person or non in-person encounters (telephone, MyChart, documentation). For the purposes of this data catalog, the variable of focus is “Smoking Status”.

Collection Date Range

Smoking status has been collected since April 2013 and is currently collected.

General Completeness Rate

Of the approximately 5.4 million patients, 1,728,749 patients (32%) reported having any value smoking status in social history. Smoking Quit Date is also populated, but only in 137,958 (3%) of the time. The status breakdown, with collection rate, is as follows:

Smoking Status selection options include:

• Current Every Day Smoker (114,566 – 2.1%)
• Current Some Day Smoker (28,547 – .5%)
• Former Smoker – (297,099 – 5.5%)
• Heavy Tobacco Smoker (3,111 – .1%)
• Light Tobacco Smoker (12,857 – .2%)
• Never Assessed (302,631 – 5.6%)
• Never Smoker – (952,636 – 17.7)
• Passive Smoke Exposure – Never Smoker – (4,274 – .1%)
• Smoker, Current Status Unknown – (1,133 – < .01%)
• Unknown if Ever Smoked – (11,915 – < .2%)

Facility Type (Inpatient or Outpatient)

Smoking status is collected at all facilities at the time of an encounter.

Other Collection Methods

For future revisions of this data catalog, alcohol use data collected from flowsheets, questionnaires, and clinical notes will be included.

Social Support

Lack of Social Support (Social Isolation) or At Risk for Social Isolation

Referred to patients in deteriorated aloneness state with lack of interaction with others, feel lonely, detached, and isolated with no help and support system. They have restricted social participation and can’t maintain social relationships. They are unable to communicate with others and have difficulty visiting friends, attending clubs, meetings, and going to parties. They might be the target of perceived adverse discrimination or persecution.

Collection Methods

In order to determine the population and distribution of patients lacking social support, several collection methods were utilized.

Diagnosis Codes (ICD-10) *

Searching Epic for patients having ICD-10 coded diagnoses on the problem list, billing codes, or recorded at the time of an encounter yielded the following results:

* Patients with ICD-9 coded diagnoses are also included in this query. The timeframe is January 1, 2003 through May 31, 2018.

Clinical Notes

Using Natural Language Processing (NLP) techniques, keyword phrases indicating specific SDH were extracted from 23 million clinic notes authored by specific provider types between July 1, 2016 and May 31, 2018. The notes represent 1,188,202 unique patients and 9,066,508 unique encounters.³

We pre-processed the notes to mark the line boundaries and remove non-Unicode characters. To identify notes containing SDH, we used hand-crafted linguistic patterns developed by a team of experts based on SDH descriptions in LOINC, SNOMED, ICD10, Public Health Surveys and Instruments (ACS, American Housing Survey, NHANE, etc.), phrases from the literature review and other studies, and manual tagging.

To craft the linguistic patterns, the expert team focused on three domains: Housing, Finance, and Social Support.

Findings for Social Support:

• Of the 1,188,202 unique patients 2.6% have at least one note containing mentions of social support.
• Of the 63,185 notes containing mentions of social support, about 70% were authored by physicians, 15% were authored by nurses, 13% were authored by social workers, 2% were authored by case managers and case coordinators.
• The top note types for mentions of social support are progress notes (46%), H&P notes (8%), treatment plans (6%), consults (6%), with the remaining note types dispersed among committee review, patient instructions, discharge summaries, and plan of care notes.

Questionnaires

Several questionnaires exist in Epic that capture data related to social support. Each is described briefly below, including frequency, general completeness rate, and facility type where the questionnaire answers were completed. Please note that data provided on general completeness reflects questions that were available to an Epic user and were answered.

References

1. Bazemore AW, Cottrell EK, Gold R, et al. “Community vital signs”: Incorporating geocoded social determinants into electronic records to promote patient and population health. J Am Med Inform Assoc. 2016;23(2):407-412
2. McGinnis JM, Williams-Russo P, Knickman JR. The case for more active policy attention to health promotion. Health Aff (Millwood). 2002;21(2):78-93
3. Kharrazi H, Lasser E, Yasnoff WA, Loonsk J, Advani A, Lehmann H, Chin D, Weiner JP. A proposed national research and development agenda for population health informatics: summary recommendations from a national expert workshop. J Am Med Inform Assoc. 2017; 24 (1):2-12
4. Kharrazi H, Chi W, Chang HY, Richards TM, Gallagher JM, Knudson SM, Weiner JP. Comparing population-based risk-stratification model performance using data extracted from electronic health records versus administrative claims. Med Care. 2017; 55 (8): 789-796
5. Chang HY, Richards TM, Shermock KM, Elder-Dalpoas S, Kan H, Alexander CG, Weiner JP, Kharrazi H. Evaluating the impact of prescription fill rates on risk stratification model performance. Med Care. 2017; 55 (12): 1052-1060
6. Lemke K, Gudzune KA, Kharrazi H, Weiner JP. Assessing markers from ambulatory laboratory tests for predicting high-risk patients. Am J Manag Care. 2018; 24(6): e190-e195
7. Kharrazi H, Weiner JP. A practical comparison between the predictive power of population-based risk stratification models using data from electronic health records versus administrative claims: setting a baseline for future EHR-derived risk stratification models. Med Care, 2017; 56(2), 202-203
8. Kharrazi H, Chang HY, Heins S, Weiner JP, Gudzune K. Enhancing the prediction of healthcare costs and utilization by including outpatient BMI values to diagnosis-based risk models. Med Care. 2018; 56 (12): 1042-1050
9. Kan H, Kharrazi H, Leff B, Boyd C, Davison A, Chang H-Y, Kimura J, Wu S, Anzaldi LJ, Richards T, Lasser E, Weiner JP. Defining and assessing geriatric risk and associated health care utilization among elderly patients using claims and electronic health records. Med Care. 2018; 56(3): 233-239
10. Hatef E, Searle KM, Predmore Z, Lasser EC, Kharrazi H, Nelson K, Sylling P, Curtis I, Fihn S, Weiner JP. The impact of social determinants of health on hospitalization in the Veterans Health Administration. Am J of Prev Med. In-press.
11. Kharrazi H, Wang C, Scharfstein D. Prospective EHR-based clinical trials: the challenge of missing data. J Gen Intern Med. 2014; 29 (7): 976-978
12. Weiskopf NG, Weng C. Methods and dimensions of electronic health record data quality assessment: enabling reuse for clinical research. J Am Med Inform Assoc. 2013; 20(1): 144–151
13. Kharrazi H, Gonzalez CP, Lowe KB, Huerta TR, Ford EW. Forecasting the maturation of electronic health record functions among US hospitals: retrospective analysis and predictive model. J Med Internet Res. 2018; 20(8): e10458
14. Kharrazi H, Anzaldi L, Hernandez L, Davison A, Boyd CM, Leff B, Kimura J, Weiner JP. Measuring the value of electronic health record’s free text in identification of geriatric syndromes. J Am Geriatr Soc. 2018; 66(1) 1499-1507
15. Dixon B, Kharrazi H, Lehman H. Public health and epidemiology informatics: recent research and events. Yearb Med Inform. 2015; 10 (1): 199‐206
16. Kharrazi H, Weiner JP. IT-enabled community health interventions: challenges, opportunities, and future directions. Generating Evidence & Methods to Improve Patient Outcomes (eGEMs). 2014; 2 (3): 1-9
17. Gamache R, Kharrazi H, Weiner JP. Public health and population health informatics: the bridging of big data to benefit communities. Yearb Med Inform. 2018; 27(1): 199-206
18. Dixon B, Pina J, Kharrazi H, Gharghabi F, Richards J. What’s past is prologue: a scoping review of recent public and global health informatics literature. Online J Public Health Inform. 2015; 7 (2) e1‐31
19. Hatef E, Lasser EC, Kharrazi H, Perman C, Montgomery R, Weiner JP. A population health measurement framework: evidence-based metrics for assessing community-level population health in the global budget context. Popul Health Manag. 2017; 21(4): 261-270
20. Hatef E, Kharrazi H, VanBaak E, Falcone M, Ferris L, Mertz K, Perman C, Bauman A, Lasser EC, Weiner JP. A state-wide health IT infrastructure for population health: building a community-wide electronic platform for Maryland’s all-payer global budget. Online J Public Health Inform. 2017; 9(3): e195
21. Kharrazi H, Horrocks D, Weiner JP. Use of HIEs for value‐based care delivery: a case study of Maryland’s HIE. In Dixon B (Ed.) Health Information Exchange: Navigating and Managing a Network of Health Information Systems. 2016; 313-332. Cambridge, MA: Academic Press Elsevier

Notes

³Disclaimer about NLP findings: The reported NLP findings are based on the occurrences/mentions of specific linguistic patterns (aka, keywords or key phrases) within a given clinical note. At this stage of the project, we did not perform a manual quality check but we think occurrences of variable-specific patterns could give us a good proxy of the variable in question.