1. 1 Public Reporting of Long Term Care Quality: The US Experience Vincent Mor, Ph.D. Brown University

2. 2 Background Long history of scandals regarding long term care quality, particularly nursing homes While preference for and supply of “community based” alternatives have grown in US, all acknowledge residentially based long term care must be part of any system Home Health less scrutinized but many worry about care adequacy since hard to inspect

3. 3 Background Institute of Medicine Report in 1987 served as basis for nursing home reform also adopted by home care Uniform Resident Assessment Instrument created in 1991 and became the basis for the creation of performance measures designed to stimulate quality competition through public reporting Home Health Outcome and Assessment Information Set (OASIS) emerged independently

4. 4 Background (cont.) Using RAI Nursing Home Quality Measures tested, revised and published as “Nursing Home Compare” since 2002 More recent efforts to create composite measure incorporating Inspection results, Staffing Levels and Quality Measures have been widely promulgated Home Health Quality Measures developed and tested and published as Home Health Compare since 2004

5. 5 Purpose Summarize US Experience with Development of Long Term Care Quality Measures Review Conceptual and Technical Issues Facing the Construction of Long Term Care Quality Measures Review Literature on Effects of Public Reporting of Quality Measures in Long Term Care

6. 6 The Nursing Home Resident Assessment Instrument (RAI) 1986 Institute of Medicine Report on Nursing Home Quality Recommended a Uniform RAI to Guide Care Planning --MDS OBRA ‘87 Contained Nursing Home Reform Act Including RAI Requirement A 300 Item, Multi-Dimensional RAI Tested for 2 Years Mandated Implementation in 1991

7. 7 Clinical Planning Basis of the MDS Assessment Profile in Given Domain “Triggers” Potential “Risk” Status Resident Assessment Protocol Reviewed to Determine Presence of Problem or High Risk of Problem Care Planning and Treatment Directed to the Problem Data Quality Contingent upon conduct of Clinical Care Planning Process

8. 8 MDS Background MDS Version 2.0 Introduced in 1996 Admission, Short Term and Quarterly Reassessments done on all Residents Inter-State Variation with some requiring additional data Since 1998 all MDS records are computerized and submitted to Centers for Medicare & Medicaid

9. 9

10. 10

11. 11 MDS: Putting Practice into Research

12. 12 CMS Quality Measures “The quality measures, developed under CMS contract to Abt Associates and a research team led by Drs. John Morris and Vince Mor, have been validated and are based on the best research currently available. These quality measures meet four criteria. They are important to consumers, are accurate (reliable, valid and risk adjusted), can be used to show ways in which facilities are different from one another, and can be influenced by the provision of high quality care by nursing home staff.” CMS Web Site

13. 13 CMS Quality Measures - Long Term Percent of Long-Stay Residents Given Influenza Vaccination During the Flu Season Percent of Long-Stay Residents Given Pneumococcal Vaccination Percent of Residents Whose Need for Help With Daily Activities Has Increased Percent of Residents Who Have Moderate to Severe Pain Percent of High-Risk Residents Who Have Pressure Sores Percent of Low-Risk Residents Who Have Pressure Sores Percent of Residents Who Were Physically Restrained Percent of Residents Who are More Depressed or Anxious (Looks back 30 days) Percent of Low-Risk Residents Who Lose Control of Their Bowels or Bladder Percent of Residents Who Have/Had a Catheter Inserted and Left in Their Bladder Percent of Residents Who Spent Most of Their Time in Bed or in a Chair Percent of Residents Whose Ability to Move in and Around Their Room Got Worse Percent of Residents with a Urinary Tract Infection (Looks back 30 days) Percent of Residents Who Lose Too Much Weight (Looks back 30 days)

14. 14 Physical Functioning- October/December 2009 State ADL Worse Bed Bound Move Worse Decline in ROM National14.9% 4.7% 14.7% 6.6% AK17.4% 5.3% 14.6% 10.9% AL11.6% 6.5% 11.7% 5.4% AR14.2% 4.2% 12.6% 5.5% AZ13.9% 4.9% 14.6% 5.5% CA10.4% 7.1% 11.6% 6.2% CO15.3% 3.0% 14.9% 6.4% CT15.4% 2.5% 16.3% 4.8% DC13.4% 1.7% 13.2% 5.4% DE13.8% 4.1% 15.5% 8.1% FL12.7% 4.6% 12.5% 5.2%

15. 15 Psychotropic Drug Use- October/December 2009 State Anti- Psychotics Overall Anti- Psychotics LOW Risk Anti-Anxiety Agents National18.6% 15.6% 23.1% AK11.2% 4.7% 21.5% AL15.9% 14.0% 27.2% AR17.9% 15.6% 21.1% AZ19.2% 15.8% 21.5% CA16.8% 14.0% 20.4% CO18.6% 15.1% 18.1% CT23.7% 21.2% 22.7% DC13.6% 12.4% 13.4% DE20.2% 17.8% 23.3% FL12.2% 10.1% 27.5%

16. 16 CMS Quality Measures – Short stay Percent of Short-Stay Residents Given Influenza Vaccination During the Flu Season Percent of Short-Stay Residents Who Were Assessed and Given Pneumococcal Vaccination Percent of Short-Stay Residents With Delirium Percent of Short-Stay Residents in Moderate to Severe Pain Percent of Short-Stay Residents With Pressure Sores

17. 17 Home Health Quality Measurement OASIS began as a cooperative effort between home health agencies and researchers to develop simple “outcome” measures to track patients’ rate of improvement while in care University of Colorado researchers worked with large Visiting Nurse Services to develop and test CMS then funded multiple large demonstrations to implement the tool and use for quality measurement and case-mix reimbursement

18. 18 Outcome Based Quality Improvement Distinct measures of change in patient functioning, resolution of symptoms and ability to manage independently collected at the start and “end” of care (OR every 60 days) Most Medicare home health is short term Measures tested and revised with extensive case mix adjustment to allow for comparison across agencies and states

19. 19 Risk-adjusted Home Health Outcome Report for Improvement of Activities of Daily Living EXAMPLE: Percent of Patients in Home Health Care whose ability to [Groom, Bathe, Dress Upper and Dress Lower Body] themselves improves between start of care and discharge

20. 20 CMS OASIS Report – 2009 Rates of Improvement in ADL StateGrooming Upper Dressing Lower Dressing Bathing Alabama71.0% 73.0% 75.0% 68.0% Alaska67.0% 66.0% 56.0% 59.0% Arizona67.0% 69.0% 67.0% 64.0% Arkansas68.0% 69.0% 70.0% 65.0% California71.0% 72.0% 70.0% 67.0% Colorado70.0% 71.0% 70.0% 64.0% Connecticut69.0% 68.0% 62.0% Delaware68.0% 69.0% 70.0% 62.0% District of Columbia75.0% 79.0% 72.0% Florida69.0% 68.0% 66.0% Georgia71.0% 73.0% 74.0% 67.0%

21. 21 Risk-adjusted Home Health Outcome Report for Utilization Outcomes Percent of patients who have received emergency care prior to or at the time of discharge from home health care. Percent of patients who are discharged from home health care and remain in the community Percent of patients who are admitted to an acute care hospital for at least 24 hours while a home health care patient.

22. 22 Risk-adjusted Home Health Outcome Report State Any Emergent Care Discharged to Community Acute Care Hospital Alabama23.0% 64.0% 33.0% Alaska20.0% 71.0% 25.0% Arizona25.0% 67.0% 29.0% Arkansas24.0% 64.0% 32.0% California18.0% 72.0% 25.0% Colorado23.0% 69.0% 26.0% Connecticut27.0% 65.0% 32.0% Delaware20.0% 70.0% 26.0% District of Columbia21.0% 71.0% 26.0% Florida18.0% 70.0% 26.0% Georgia22.0% 68.0% 29.0%

23. 23 Conceptual Issues Inherent in Applying Quality Indicators Requires “shared” interpretation of Quality Assumes all Providers have same goals Assumes Measured Quality Domains are Important Indicators are NOT Quality per se, BUT often used as evidence in and of themselves Assumes Facilities Accountable for most of the variation in the Indicator (e.g. outcomes) Assumes Facilities Know how to Change Practice

24. 24 Technical Issues That Can Compromise Validity of QI’s Reliability & Validity of the data Multi-dimensionality of Quality & Indicators Stability of Estimates Sensitive to Sample Size Ranks can Overestimate Differences Patient Level Risk Adjustment Complex Differences in Assessment Practices Influence QI Scores & Comparisons

25. 25 Reliability Studies: NH 219 of 462 (47.4%) facilities approached chose to participate in full study (52.4% for HB and 45.6% for non-HB); Non-participants were more likely to be for-profit, less well staffed and with more regulatory deficiencies 5758 patients (ave. 27.5/facility) included in reliability analyses; 119 patients assessed twice by research nurses Patients resemble traditional US nursing home patient

26. 26 Reliability of “Gold Standard” Nurses Of 100 items, only 3 didn’t reach Kappa>.4 50%+ items had Kappa >.75 Pct. Agreement high even for ordinal items with variance Item% AgreeKappa DNR91%.83 Memory88%.63 Decisions97%.89 Understood96%.82 Understand96%.80 Fears97%.76 Wander99%.85 Walk95%.86 Pain Fx.93%.78

27. 27 Reliability of Facility RNs to “Gold Standard” Of the 100 data items 28 had Kappa.75 Worst Kappa items were rare binary items like “end stage”, didn’t use toilet, recurrent lung aspirations, etc. ADLs and other Functioning items had Kappa values above.75

28. 28 Reliability of Constructed Quality Indicators: NH Quality Indicators are composites of several RAI items; a definition of the denominator and of the conditions required to meet the QI definition The inter-rater reliability of a QI is a function of the reliability of all the component items defining the algorithm

29. 29 Prevalence and Inter-Rater Agreement and Reliability of Selected Facility Quality Indicators [N=209 homes] Avg. QI Prev rate Facility Ave SD of QI Prev rate Ave Kappa for Items used in QI % Agree Resch & facility RNs on QI QI Kappa Behavior Problems High & Low Risk Combined.20.10.7189.8.61 Little no activities.12.12.2865.3.23 Catheterized.07.05.7192.5.67 Incontinence.62.13.8891.4.78 Urinary Tract Infection.08.05.5389.1.45 Tube Feeding.08.05.7398.1.83 Inadequate Pain Management.11.08.8586.5.87

30. 30 Facility QI Reliability Variation: Bladder/Bowel Incontinence

31. 31 Facility QI Reliability Variation: Inadequate Pain Management

32. 32 Reliability Studies: Home Health Fewer inter-rater reliability studies of OASIS More expensive to send two nurses at separate times on the same day to do the same assessment Largest Reliability Study done as part of research to develop case-mix reimbursement system ADL and other function items yield high levels of reliability; symptoms achieve “ok” reliability

33. 33 Selected Inter-Rater Reliability Results from OASIS test Signs & Symptoms Sampl e Size Percent Agreement Kappa 1. Diarrhea30493.4%0.44 2. Difficulty urinating or >=3x/night30491.5%0.45 3. Fever30496.7%0.63 4. Vomiting30497.4%0.49 5. Chest Pain30495.4%0.51 6. Constipation in 4 of last 7 days30492.1%0.53 7. Dizziness or lightheadedness30489.1%0.46 8. Edema30481.3%0.50 9. Delusions30499.0%0.66 10. Hallucinations30498.4%0.44

34. 34 OASIS Reliability Results: Function VariableSample Size Percent Agreement Kappa Grooming: Current ability to tend to personal hygiene needs 30474.7%0.83 Dressing: Current ability to dress upper body with or without dressing aids 30471.1%0.83 Dressing: Current ability to dress lower body with or without dressing aids 30477.0%0.85 Bathing: Current ability to wash entire body30464.8%0.80 Toileting: Current ability to get to and from the toilet or bedside commode 30482.6%0.86 Transferring: Current ability to move from bed to chair, on/off toilet or commode, tub, … 30474.3%0.88 Ambulation/Locomotion: Current ability to safely walk, use a wheelchair… 30477.6%0.87

35. 35 Validity of the Data & Measures Validity of the data shown by the extent to which items and measures behave as expected relative to “gold standard” variables or “hard” outcomes Compared MDS diagnoses to Hospital discharge diagnoses Looked at MDS predictors of survival Related to MDS measures to research scales

36. 36 MDS vs. CMS Hospital diagnoses Neurological Cerebrovascular disorders (ICD-9: 432, 434, 436, 437)  PPV = 0.73 Parkinson’s disease (ICD-9: 332)  PPV = 0.86 Alzheimer’s disease (ICD-9: 331)  PPV = 0.68 Brain degeneration (ICD-9: 331.0, 331.2, 331.7, 331.9)  PPV = 0.84

37. 37 Men (CPS 0-1) Women (CPS 2-4) Months One Year Survival by Gender & Cognition Level

38. 38

39. 39 Construct Validity: Cognitive Performance Scale & Correlates Cognitive Performance Scale (CPS) Derived from 5 MDS Items Strong (>.85) Correlation with MMSE High Kappa with Global Deterioration Scale (.76) Percent Patients with Dementia Increases as CPS Declines MDS Communication Correlated (.85) with MMSE ADL, CPS Symptoms & Select Diagnoses Related to Survival

40. 40 Sample Size and QI Stability  Providers and Consumers want QI to reflect not just what WAS but what WILL BE; SO  QI stability is desired  QI must be based upon minimum # observations  Correlation between quarters among QIs varies  Correlation among prevalence based QIs is high because same individuals assessed each quarter  Correlation between quarters among incidence and change based QIs lower and VERY sensitive to sample size

41. 41 Residents’ Expected Rates of Change on Quality Indicators  Over 90 day period 77.1% of residents still in facility do not change on ADL, 14.7% decline and 8.2% improve.  Over 12 months 58% of residents in home don’t change and 30.2% decline.  Similar pattern for Communication, Cognition and individual ADL items  Means that rates of decline are low and many residents are needed to estimate a home’s rate of ADL decline with confidence.

42. 42 Estimated Sample Size for Change Facility Size Number Residents Decline Estimate Residents Expected to Decline 20 th Pctile Expected Residents Declining 80 th Pctile Expected Residents Declining 20 Beds 12 1 <1 1 30 Beds 16 1 <1 3 50 Beds 28 2 1 4 80 Beds 45 4 2 6 100 Beds 56 5 2 7 150 Beds 83 7 4 11 200 Beds 117 9 5 14

43. 43 Long Term Predictability of Quality

44. 44 Quality Fluctuation: Seasonality

45. 45 Transforming QI Scores into Ranks  Many QI score distributions are skewed; many facilities with little or no problem and few facilities with many residents experiencing the problem.  Median facility might be very similar to the “best” (the one with fewest problems)  Transforming to ranks means saying there is a difference between the 10 th and 40 th percentile when there is little difference

46. 46 Pressure Ulcer Prevalence Facility Distribution: Meaning of Ranks

47. 47 Persistent Pain: Median Ranks Anti-psychotics: Median Ranks Anti-psychotics: Median Ranks Variability in Ranking Distributions

48. 48 Complexity of Determining Appropriate Risk Adjustment Risk Factors May not be Measured Independent of the Provider (tx) Effect Potential for Over Adjustment as Great as Under Adjustment How to Adjust for Socio-Economic Differences Known to Affect Health Behavior or Clinical Characteristics ( e.g. PU not “seen” on African American NH pts until at Stage 2 OR Pain Harder to “see” in Cognitively Impaired & Oldest pts)

49. 49 Risk Adjustment Complexity

50. 50 Why Adjust QIs Facilities should be compared on ‘level playing field’, acknowledging differences in  Types of residents admitted  Ability to ameliorate clinical characteristics thought to predispose to poor outcomes irrespective of care quality  Variability in measurement acumen of assessors

51. 51 Average Admission Prevalence of Pressure Ulcers Across All States, 1999

52. 52

53. 53

54. 54 Multi-dimensionality of QIs Consumers want to know “Best” nursing home & Regulators want to know where to focus their survey energies & Purchasers want to buy best. If Quality is multi-dimensional no such thing as the “best”; most valuable dimension is a preference and will be individualized Combining QIs that aren’t highly correlated may mask differences between facilities on important individual QIs

55. 55 Does Poor Performance on One Measure Mean NF is Poor? Average Correlation Among QIs is Low; Anti-Psychotics and Restraints Correlated.04 What is a “Good” Home if QIs not Related? Can Performance Measures Help Pick Good Homes? Are Some Measures More Meaningful? Should Users of Performance Measures Select the Measures they Value Most?

56. 56 Summary Results of Factoring Functional Decline Mood/ Behavior Pressure Ulcers Treatment & Condition [No Factor] Worsening BladderPoor Mood State [Prevalence] Worsening Pressure Ulcers Prevalent Catheter Worsening BowelWorsening MoodPrevalent Pressure Ulcer (Hi Risk) Prevalent Restraint ADL DeclinePoor Mood w/o Anti- depressants Prevalent Pressure Ulcer (Lo Risk) Prevalent Anti- Hypnotic Use Mobility DeclineBehavior Problems [Prevalence] Prevalent Anti- Psychotic Use Cognitive DeclineWorsening BehaviorWeight Loss Communication Decline Worsening Relationships Falls Worsening Pain

57. 57 Regression Modeling Results  Results of relating each QI to all others revealed very low R 2 for all Treatment & Conditions  While R 2 higher for QIs within other factors, many conceptually unrelated QIs found to weakly predict other QI  Many QIs “load” (related to) on multiple factors  QI “type” (e.g. prevalence, longitudinal, change) as influential as QI content in factor  Factor structure sensitive to which QIs included  Many QIs totally uncorrelated with others

58. 58 Provisional Test of Combining Unlike Quality Indicators Use 1999 MDS 2.0 from OH, NY & CA Create Risk and Admission adjusted QIs for Pressure Ulcers, Anti-Psychotic Use and Pain Correlate Measures: PU & Pain<.05; PU & Anti- Psych = -.15; Pain & Anti-Psych =.16 Only 13% of facilities in bottom half on all 3 QI’s; 5% if use bottom third on all measures

59. 59 Public Reporting of Quality NURSING HOME COMPARE allows consumers and advocates to identify facilities in their geographic area and to select using a “Five Star” global rating OR based upon global domains OR specific measures. HOME HEALTH COMPARE allows consumers and advocates to identify agencies in geographic area and presents detail of many different Quality Indicators

60. 60

61. 61

62. 62

63. 63

64. 64

65. 65

66. 66 Effect of Public Reporting Research to date all done on nursing homes Broken down by market served; long stay residential vs. short stay, post-acute First wave of studies surveyed administrators to find out how they were responding More recent studies use MDS data to examine changes in outcomes and admission patterns

67. 67 Facility Response to Reporting Castle (2005) initially found Administrators were skeptical and unconvinced that reporting mattered Zinn & colleagues (2005) surveyed leaders and found they were aware of their scores and those of closest competitors; concluded spurred quality improvement Castle (2007) concurred in a separate survey that more competitive markets affected response

68. 68 Reporting Improve Quality? Werner & colleagues (2009) found significant improvement in BOTH measured and unmeasured quality measures following public reporting – BUT general improvement trend Mukamel et al (2007) looked carefully at initial response relative to prior quality patterns and also found improvement on most but not all measures Werner, et al, 2010 also found improvement in post-acute quality scores

69. 69 Reporting Alter Admissions? Werner & Colleagues have examined whether facilities with worse quality scores in competitive markets manifest reductions in admissions Very complicated; must infer from the data why someone entering a facility; should affect those entering to stay more, but hard to know who However, evidence suggest small but significant changes in referral patterns favoring better quality

70. 70 Summary Public Reporting of long term care providers’ quality performance is possible; All measures are flawed, but no more than acute and ambulatory care Pre-requisite is to have uniform data collected with relevant clinical detail AND should be able to be audited with penalties to minimize bad data

71. 71 Summary (cont.) Constructing quality measures can be complex Sample size, seasonality, risk adjustment are all important to assure the “fairness” of the system Like case mix reimbursement, don’t want incentives for providers to limit access to sickest Still at the infancy of understanding how consumers & advocates use the data

72. 72 Issues for the Future Preferable to have common items, measures and metrics across different types of long term care options, technically AND for consumers Challenge of Creating Composite Scores consumers want that are technically less sensitive than domain specific measures However, Movement to “Pay for Performance” requires we develop a solution