Patient Safety Indicators’ Experiences in France and in Switzerland Jean-Marie Januel, PhD, MPH, RN Senior Researcher University hospital of Lausanne, Institute of Social and Preventive Medicine, Lausanne, Switzerland Health Services Research Seminar Series UC Davis, Center for Healthcare Policy and Research Sacramento, CA, US – May 9, 2013 IUMSP Institut universitaire de médecine sociale et préventive, Lausanne 1 Disclosure Statement Speaker’s Verbal Disclosure Statement: Have you (or your spouse/partner) had a personal financial relationship in the last 12 months with the manufacturer of the products or services that will be discussed in this CME activity? ___ Yes _X_ No (If yes, please state disclosures and resolutions) Educational objectives To describe the French experience of PSI (focusing on feasibility, validity and measurement variations between providers) To describe the Swiss experience of PSI (focusing on the use of PSI for assessing the impact of DRG implementation in all Swiss hospitals) To discuss about an appropriate reference standard for comparing Patient Safety Indicators (example provided by a pilot study to compare the PSI 12 – postoperative pulmonary embolism and deep vein thrombosis – in hospitalized patients undergoing hip arthroplasty between France and Switzerland) BACKGROUND Hypothesis to Develop New Indicators To use Administrative Databases for assessing adverse events related to healthcare that could be preventable McDonald K, Romano P, Geppert J, Davies SM, Duncan BW, Shojania KG, et al. Measures of Patient Safety Based on Hospital Administrative Data - The Patient Safety Indicators. Technical Review 5. AHRQ Publication No. 02-0038 . Rockville, MD: Agency for Healthcare Research and Quality Approach derived from the Complication Screening Program (CSP) by Iezzoni LI et al. Iezzoni LI, Foley SM, Heeren T, Daley J, Duncan CC, Fisher ES, Hughes J. A method for screening the quality of hospital care using administrative data: preliminary validation results. QRB Qual Rev Bull. 1992;18(11):361-371. PSI Algorithm Model ICD Codes for Secondary Diagnoses related to Adverse Event Identification PSI = Population at risk, defined using DRG codes, Diagnostic codes, Procedure codes Projects 20 Patient Safety Indicators (PSI) initially developed by the Agency for Healthcare Research and Quality (AHRQ) using ICD9-CM McDonald K, et al. Measures of Patient Safety Based on Hospital Administrative Data - The Patient Safety Indicators. Technical Review 5. AHRQ Publication No. 02-0038 . Rockville, MD: Agency for Healthcare Research and Quality ICD10 adaptation of 15 PSI by the International Methodology Consortium for Coded Health Information in 2007 (www.imecchi.org) Quan H., et al. AHRQ Advances in Patient safety 2: News Directions and Alternatives Approaches. Vol 1. Assessment. Taxonomies and Measurement. Agency for Health Care Research and Quality Publication 2008. Januel JM, et al. Rev Epidemiol Sante Publique 2011;59 :341-350. Achievements for PSI Benchmark (external comparisons) o Between healthcare providers (hospitals) o Between healthcare systems (countries) Surveillance (internal comparisons) o Longitudinal studies (to assess epidemiological peak) o To assess changes in robustness relating of changes in ICD coding rules / version used The Patient Safety Indicators in the Literature 40 36** Total citations for PSI 35 33 Review / Commentary / Editorial 30 Fréquency, n Original study 26* 25 20 17 14 15 10* 10 5 6 5 1 1 2001 2002 0 2003 2004 2005 2006 2007 Year of publication *Including one citation without abstract; **including two citations without abstract. 2008 2009 2010 The French experience of PSI French Pilot Study – Objectives To explore French data using PSI algorithms (ICD10 version - IMECCHI) To assess feasibility for PSI by calculating a selection of PSI from nationwide database in 2005 and 2006, respectively To validate selected PSI using chart review PSI estimates calculation for feasibility Frequency, N PSI # Denominator Prevalence Numerator /1000 stays (σ) Incidence density /1000 days hosp. (σ) Mean Prevalence at hospital level /1000 stays (σ) 1 3’124’476 707 0.23 (0.02) 0.05 (0.00) 0.14 (0.78) 3 2’923’535 20’734 7.09 (0.10) 0.66 (0.00) 10.11 (17.06) 5 8’973’343 426 0.05 (0.00) 0.01 (0.00) 0.03 (0.11) 7 6’078’340 2’177 0.36 (0.02) 0.06 (0.00) 0.33 (2.01) 10 3’101’929 21’605 6.97 (0.09) 1.29 (0.01) 4.79 (15.36) 12 3’123’112 16’719 5.35 (0.08) 0.91 (0.01) 3.62 (10.44) 13 1’169’288 6’074 5.19 (0.13) 0.61 (0.01) 2.70 (5.10) 15 7’989’654 5’819 0.73 (0.02) 0.12 (0.00) 0.56 (1.15) 16 8’973’561 53 0.01 (0.00) 0.00 (0.00) 0.00 (0.06) 17 756’768 2’392 3.16 (0.13) 0.68 (0.01) 1.43 (7.26) 18 57’499 803 13.97 (0.96) 2.89 (0.10) 5.00 (18.10) 19 570’404 1’788 3.13 (0.15) 0.71 (0.02) 1.22 (3.08) 20 150’808 10 0.07 (0.04) 0.01 (0.00) 0.02 (0.36) Stratified analysis for PSI #12 (Postoperative PE /DVT) 120 30 2005 HMen 100 2006 HMen 2005 2005 FWomen 20 2006 FWomen 15 10 PSI /1000 discharges PSI /1000 discharges 25 60 40 5 20 0 0 [18-40[ [40-65[ [65-75[ Age Categories [75-85[ >= 85 ans 2006 80 1 [2-4[ [4-8[ [8-15[ [15-22[ [22-29[ >= 29 Lenght of Stay in Hospital (days) Estimating PPV based on three PSI To compare PSI positive cases to a retrospective chart review of medical records Between 80 and 150 inpatients stays with positive PSI at the university hospital of Lyon, France To calculate the Positive Predictive Value (PPV) To identify potential reasons for false positive cases PPV of PSI #12 (postoperative PE/DVT) University Hospital, Lyon, France (2008) True Positive Cases Positive PSI 12 Sample N Total cases 154 VPP (%) N (100.0) % 123 79.87 [CI 95%] [72.66-85.89] Sex 0,607 Men 61 (39.6) 45 73.77 [60.93-84.20] Women 93 (60.4) 78 83.87 [74.80-90.68] Type of surgery <0,001 Hip Arthroplasty* 43 (27.9) 38 88.37 [74.92-96,11] Knee Arthroplasty** 58 (37.7) 56 96.55 [88.09-99.58] Femur Fracture (Osteosynthesis) 19 (12.3) 15 78.95 [54.43-93.95] Other orthopedic procedures 13 (8.4) 8 61.54 [31.58-86,14] Digestive procedure 13 (8.4) 6 46.15 [19.22-74.07] Thorax & Pulmonary procedure 8 (5.2) 0 0.00 [0.00-36.94] Chart review Paper record only Both paper and electronic P 0,232 114 (74.0) 83 72.81 [63.67-80.72] 40 (26.0) 40 100 [91.19-100] PPV of PSI #7 (Bacteremia-related to Catheter) University Hospital, Lyon, France (2008) True Positive Cases Positive PSI 7 sample Total cases PPV N (%) N 55 100 16 % 29.09 [CI 95%] P [17.63-42.90] Sex 0.857 Men 40 72.73 12 30.00 [16.56-46.53] Women 15 27.27 4 26.67 [7.79-55.10] Hospital unit 0.445 Medicine 19 34.55 7 36.84 [16.29-61.64] Infectious diseases 14 25.45 5 35.71 [12.76-64.86] Surgery 16 29.10 2 12.50 [1.55-38.35] Clinical Nutrition 3 5.45 1 33.33 [0.84-90.57] ICU 3 5.45 1 33.33 [0.84-90.57] Chart review 0.255 Paper record only 18 32.73 4 22.22 [6.41-47.64] Electronic record only 32 58.18 9 28.13 [13.75-46.75] Both paper and electronic 5 9.09 3 60.00 [14.66-94.73] PPV of PSI #13 (postoperative sepsis) University Hospital, Lyon, France (2008) True Positive Cases Positive PSI 13 sample Total cases PPV N (%) N % [IC 95%] 81 (100) 21 25.93 [16.82-36.86] Sex P 0.351 Men 48 (59.26) 14 29.17 [16.95-44.06] Women 33 (40.74) 7 21.21 [8.98-38.91] Lenght of Stay 0.006 <20 days 40 (49.38) 5 12.50 [4.19-26.80] ≥20 days 41 (50.62) 16 39.02 [24.20-55.50] Chart review <0.001 Paper record only 27 (33.33) 14 51.85 [31.95-71.33] Electronic record only 54 (66.67) 7 12.96 [5.37-24.90] 43 (53.09) 2 4.65 [5.68-15.81] Coding rule Code « R578 » <0.001 The Swiss experience of PSI IDoC Project (Switzerland) To assess the Impact of Diagnosis related groups (DRG) implementation on patient Care and professional practice in Swiss hospitals o Overall Hospitals DRG implementation in Switzerland for inpatient stays payment on 2012, January 1st Design o 5 sub-projects based on several outcomes (Ethic; Law; Nursing sensitive; AMI; and PSI) Sub-project “PSI” o To monitor the possible effects of the generalization of DRGbased hospital reimbursement using Patient Safety Indicators (PSI) To take into account the number of SDx in Models (hierarchical) better At inpatient Level o To control differences in case mix (using categorical variable) At Hospital Level o To control differences in quality of coding o By assessing adjusted average number of SDx for each hospitals using negative binomial regression models SDx in a Previous Analysis of PSI #12 2008 2009 2010 Mean # of SDX (observed) 2.11 (2.10 – 2.12) 2.19 (2.18 – 2.19) 2.34 (2.33 – 2.35) Mean # of adjusted SDX (random effects of hospital) 2.48 (2.47 – 2.49) 2.52 (2.51 – 2.53) 2.66 (2.65 – 2.67) Between hospitals variance 0.89 (0.88 – 0.90) 0.73 (0.72 – 0.73) 0.61 (0.61 – 0.62) Mean # of adjusted SDX (random effects of hospitals + fixed effect for sex, age, los and type of hospitals) 2.66 (2.64 – 2.67) 2.66 (2.65 – 2.68) 2.78 (2.77 – 2.79) Between hospitals variance 0.74 (0.73 – 0.74) 0.62 (0.62 – 0.63) 0.55 (0.55 – 0.55) % of variance explained by fixed effects variables 16.85% 15.07% 9.84% Impact of Changes in ICD10 version on PSI 12 estimates Changes in ICD10 version occurred between 2008 and 2010 in Switzerland o 2008: ICD10-WHO o 2009: ICD10-GM2008 o 2010: ICD10-GM2010 Trend comparisons across years (2008, 2009 and 2010) using ICD-10-WHO To compare ICD-10-WHO to ICD-10-GM using: o Data from 2009 o Data from 2010 Hierarchical Logistic Regression Model for PSI #12 2008 ICD-10-WHO 2009 ICD-10-WHO 2010 ICD-10-GM2008 ICD-10-WHO ICD-10-GM2010 % postoperative VTE using the PSI #12 (observed) 0.34 (0.30 – 0.36) 0.38 (0.36 – 0.37) 0.38 (0.36 – 0.37) 0.38 (0.37 – 0.40) 0.38 (0.37 – 0.40) % postoperative VTE using the PSI #12 (random effect of hospitals) 0.34 (0.34 – 0.34 ) 0.37 (0.37 – 0.37) 0.37 (0.37 – 0.37) 0.38 (0.38 – 0.38) 0.38 (0.38 – 0.38) Between hospitals variance 1.23 (0.84 – 1.80) 1.06 (0.72 – 1.56) 1.06 (0.72 – 1.56) 1.20 (0.82 – 1.75) 1.20 (0.82 – 1.75) ICC 0.27 (0.20 – 0.35) 0.24 (0.18 – 0.32) 0.24 (0.18 – 0.32) 0.27 (0.20 – 0.35) 0.27 (0.20 – 0.35) MOR 2.88 C-statistic 0.720 (0.709 – 0.730) 2.67 0.719 (0.709 – 0.729) 2.67 0.719 (0.709 – 0.729) 2.84 0.721 (0.711 – 0.731) 2.84 0.721 (0.711 – 0.731) To Develop a New “PSI 13” To refine definition from Postoperative Sepsis to Postoperative Hospital-Acquired Infections To develop a New Algorithm To compare New Algorithm to PSI 13 (AHRQ) To repeat models using data from 2008 to 2010, with respect to changes in ICD-10 version used each year in Switzerland Approach for New Algorithm (ICD10) Infections codes at SDx Fields (3 scales indicator) Criteria for Severity using codes at SDx Fields Bacteremia / Blood Stream Infection (A49, B95, B96) Pneumonia (J13 to J18) Nosocomial pneumonia (U6900) Sepsis (A40, A41, A419 according ICD10 version, and B377) Septic shock (A419 or R572, according the ICD10 version) SIRS (R65) Acute Organ Dysfunction (AOD) – Inspired from Angus et al. (2001), Martin et al. (2003), and Sundararajan et al. (2005) Healthcare-Related / Nosocomial criteria at SDx Fields T-Codes for Healthcarerelated infections (T802, T814, T826, T827, T835, T836, T845, T846, T847, T857, T874) Y-Codes for “Nosocomial” (Y95, Y69, Y62 according the ICD10 version) New Algorithm vs. PSI #13 (AHRQ) 2008 (ICD10-WHO) N PSI #13 (AHRQ algorithm) (%) Nosocomial criteria (%) 2009 (ICD10-GM2008) 2010 (ICD10-GM2010) N N (%) Nosocomial criteria (%) (%) Nosocomial criteria (%)) 1 727 (0.50) (1.91) 1 835 (0.52) (13.51) 2 582 (0.54) (14.56) 18 235 (5.23) (0.67) 22 077 (5.90) (4.96) 23 646 (4.95) (6.71) Sepsis / Severe Sepsis 809 (0.23) (0.26) 1 327 (0.35) (13.79) 2 617 (0.74) (18.27) Septic Shock 637 (0.18) (2.35) 800 (0.21) (11.88) 913 (0.19) (15.66) New Algorithm (including 3 levels of severity) Bacteremia / Pneumonia PSI rate “high” or “low ”? A pilot study to compare Swiss and French data for PSI #12 (Postoperative PE/DVT) To which Benchmark / Reference Standard Compare PSI 12 (postoperative PE/DVT)? The Hypothesis o Need a BENCHMARK for assessing comparisons across clusters (e.g., hospitals, countries…) in studies for healthcare quality improvement Possible benchmarks / reference standards o Risk-adjusted observed vs. expected rate (Funnel plots) o Benchmark vs. « best in class » (Forest plots) o Zero event o Evidence-based data / information Development of an evidence-based benchmark for PSI 12 (postoperative PE/DVT) Proposal: to compare actual values to an evidence-based reference standard Three step methodology o To develop a “reference standard” using the baseline risk of Postoperative PE/DVT occurring in hospitalized patients undergoing hip arthroplasty, under appropriate prophylaxis, using a systematic review with meta-analysis o To estimate adjusted occurrence rates of PSI 12 in patients undergoing hip arthroplasty (pilot study using data from Switzerland and France) o To compare these rates against the “reference standard” as a benchmark (or a target) developed using the meta-analysis Systematic Review (Januel JM, et al. JAMA 2012;307(3): 294-303) Original RCT and observational studies published from 1996 to 2011 (PubMed-Medline, EMBASE, Cochrane) Adult patients undergoing hip arthroplasty with appropriate VTE prophylaxis (updated guidelines during the corresponding period) Proportions of symptomatic VTE occurring between arthoplasty and hospital discharge using pooled occurrence rate with random effects GRADE1 method for assessing quality of included studies and for evaluating the evidence-basis of our systematic review results (1) GRADE working group. www.gradeworkinggroup.org Systematic Review – RESULTS 27 studies 21’369 adult patients undergoing hip arthroplasty 38 prophylactic treatment subgroups 58 to 70 years old on average Follow-up after surgery ranged from 8 to 17 days Pooled VTE 0.53% (95% CI, 0.35% to 0.70%) Heterogeneity: I2=49.4% P<0.001 Pooled estimates by type of prophylaxis % (95% CI) I² P LWMH (OS) 0.83 (0.19 – 1.48) 67.3% 0.230 LWMH (RCT) 0.51 (0.26 – 0.76) 45.4% 0.010 Direct Inhibitor, Factors IIa/Xa (RCT) 0.31 (0.03 – 0.59) 32.8% 0.070 Indirect Inhibitor, Factors IIa/Xa (RCT) 0.68 (0.26 – 0.97) 0.0% 0.380 0.53 (0.35 – 0.70) 49.4% <0.001 TOTAL PSI 12 estimates using Swiss / French Data Routine Data (ICD10) from overall Swiss / French hospitals (3 consecutives years) Hierarchical Two-Level Logistic Regression Model for estimating PSI #12 outcomes o Level-1 = Inpatients o Level-2 = Hospitals Risk-adjustment Model Fixed effect variables at each level of the model: o Inpatients Case mix (Sex, Age, Death, # of Secondary Diagnoses coded, Selection of comorbidities from Charlson and Elixhauser indices) Usual practice for detecting DVT (e.g., lower extremities ultrasound before hospital discharge) o Hospitals Average number of secondary diagnoses coded (adjusted on sex, age and length of stay using negative binomial regression models) Number of hip arthroplasty procedures Adjusted PSI 12 using Two-Level Logistic Regression Models France Switzerland 2008 2009 2010 2006 2007 2008 0.35% 0.34% 0.35% 1,36% 1,31% 1,38% (0.31 – 0.40) (0.30 – 0.38) (0.32 – 0.39) (1.17 – 1.55) (1.15 – 1.47) (1.22 – 1.55) 0.266 (0.215) 0.256 (0.217) 0.090 (0.172) 1.064 (0.104) 0.812 0.782 0.842 0.849 (0.766 – 0.857) (0.729 – 0.835) (0.801 – 0.883) VTE (PSI #12) Expected Variance (SE) C-statistic 1.052 (0.102) 1.026 (0.099) 0.850 0.857 (0.841 – 0.857) (0.842 – 0.858) (0.849 – 0.865 ) Factors associated to VTE Switzerland o Some comorbidities are different across years (lower frequency of some comorbidities, thus not included in models) France o The systematic use of lower extremity ultrasound for screening DVT before discharge in almost 20% of inpatients Comparisons of Swiss and France (PSI 12) against reference standard (meta-analysis) Stratified on Length of Stay Displayed on a same figure including: o Adjusted proportions of VTE using the PSI #12 (data from Switzerland and France) o Pooled proportion of VTE occurring before hospital discharge, using metaanalysis (LMWH prophylaxis according to ENDORSE Study) 3 Proportion of VTE (%) Comparisons stratified on length of stay 2 Switzerland 76.4% France 74.1% 1 LMWH = 0.58% (0.35% - 0.81%) 0 < 8 days [8 - 18[ ≥ 18 days Expected values (from Models) using Swiss and French Nationwide Data, stratified by Length of Stay and by Year International Perspectives Enlargement of the Swiss-French Pilot Study to several other country (Canada, Germany, Australia, New-Zealand, South-Korea, USA …) Expected issues o Overall nationwide data or 10% (or more) representative sample? o Which countries do not have sufficient data quality for assessing practice using procedure codes? o Which other potential reasons for explaining differences between countries Conclusions Facts … We are able to compare all outcomes that could be measured for HSR… But we are only able to interpret measures using the same definition / the same metrology / the same data quality… Or, for which we are able to control potential biases… A Questionnaire for Conducting Studies to Assess Quality of Healthcare Which variations across and between providers (hospitals, health systems as countries)? What are the sources of these variations (Clinical definition, Algorithm codes and Coding rules, Data quality, Case-mix, Quality of care, Random effects)? How to control these potential biases (models, risk-adjustment variables, interpretation)? Back to Basic… Conditions for assessing healthcare quality using patient safety indicators based on the relation between a Process, an Outcome, and a Structure o Before hospital discharge (AEs related to healthcare in hospitals / potentially no sufficient consistency and accuracy for data after discharge) o Outcome compared to “state of the art” practices (guidelines, recommendations, systematic review with meta-analysis) o Results should be interpretable for decision making Approach for Modeling Comparisons between Providers / Health Systems Take into account (adjustment, stratification) o Difference in practice (use of prophylaxis to prevent VTE, see ENDORSE Study) o Difference in measurement (use of different methods for assessing a diagnosis) o Difference in coding diagnoses (Coding rule for Major Diagnosis, # of Secondary Diagnoses coded) o Difference between Health Systems and Hospitalization habits (average length of stay between countries) Giving Sense to our Findings "We have to assess the quality of our measures and make sure that they make sense to clinicians, make sure providers can act upon them and that we account for variation in how sick the patients are." Prof. Patrick S. Romano From CHPR Website Jean-Marie.Januel@chuv.ch IUMSP Institut universitaire de médecine sociale et préventive, Lausanne 46 Inpatients Characteristics VARIABLES Switzerland * France ** P P YEARS (Strate in analyses), N (%) 1= 2008* / 2008** 20 685 (32.48) 137 136 (32.81) 2= 2009* / 2007** 21 228 (33.33) 139 243 (33.32) 3= 2010* / 2006** 21 774 (34.19) 141 559 (33.87) HOSPITALS (Level-2), N / Average vol. of hip arthroplasty procedures [min-max] Year #1 137 217 [1 – 887] 728 219 [1 – 3 643] Year #2 134 228 [1 – 811] 722 221 [1 – 3 651] Year #3 128 238 [1 – 825] 699 239 [1 – 3 917] INPATIENTS (Level-1) Sex (Women), N (%) 35 305 (55.44) 0,334 252 687 (60.46) <0.001 Age, Mean (SD) 70,06 (12.36) 0,558 72,00 (12.56) <0.001 Lenght of Stay, Mean (SD) 11,85 (7.04) <0,001 11,78 (7.25) <0.001 Secondary Diag., Mean nb. (SD) 2,29 (2.71) <0,001 2,33 (2.59) <0.001 Death, N (%) 669 (1.05) 0,836 5 167 (1.24) 0.205 236 (0.37) 0,946 5 898 (1.41) 0.264 PSI #12, VTE (both PE+DVT), N (%) 2008 2009 0+ 50 0[ -5 0 0[ [4 50 -4 5 0[ [4 00 -4 0 0[ [3 50 -3 5 0[ [3 00 -3 0 0[ 50 -2 5 0[ 010 <5 [5 50 -5 0 50 -4 5 [4 [4 00 -4 0 50 -3 5 [3 00 -3 0 [3 [2 50 -2 5 00 -2 0 [2 50 -1 5 [1 0[ 00 <5 010 [5 [1 Samplesize from hospitals [2 0.0 0[ 0.0 00 0.5 -2 0 0.5 [2 1.0 0[ 1.0 50 1.5 [1 1.5 -1 5 2.0 00 2.0 [1 2.5 0 2.5 0+ 3.0 0[ 3.0 0[ 3.5 0[ 3.5 0[ 4.0 0[ 4.0 0[ 4.5 0[ 4.5 0[ 5.0 0 5.0 Adjusted # of SDx (hospital level) Adjusted Average # of SDx at hospital Level using Negative Binomial Regression Samplesize from hospitals 2010 2006 2007 2008