Healthcare Reliability Science

Transcription

1 Healthcare Relablty Scence Delberate Desgn James C. Benneyan, PhD, Drector Healthcare Systems Engneerng Insttute CMS Innovaton Healthcare Systems Engneerng Center NSF Center for Health Organzaton Transformaton Northeastern Unversty, Boston MA

2 Systems Engneerng domans 3 Core Areas Area Focus Typcal methods System/process desgn Desgn (re-desgn) new system to meet certan functonal crtera Systems engneerng Desgn for X Busness process re-engneer 2 System/process mprovement Improve performance of exstng system Qualty mprovement Lean, sx sgma, PDSA Statstcal analyss Model-based mprovement 3 System/process optmzaton Optmze performance exstng or new system Mathematcal optmzaton Computer smulaton Machne learnng Analytcs X = Manufacturng, ds-assembly, aglty, relablty, qualty, robustness, etc 2 Healthcare Systems Engneerng Insttute Northeastern Unversty 202

3 Healthcare Systems Engneerng Insttute 3 Northeastern Unversty 202

4 Desgn for Relablty Healthcare Systems Engneerng Insttute 4 Northeastern Unversty 202

5 Outlne / Learnng objectves. Context and defntons 2. Relablty scence (delberate desgn) 3. Evdence-based bundles 4. Actual relablty engneerng 5. Dscusson Healthcare Systems Engneerng Insttute 5 Northeastern Unversty 202

6 . context and defntons Healthcare Systems Engneerng Insttute 7 Northeastern Unversty 202

7 Types & defntons of relablty Applcaton Area Classc engneerngndustry relablty Measurement and nstrument analyss (metrology) Assessment scale or survey relablty Healthcare process relablty Focus, defnton Ablty of a system or component to perform ts requred functons under stated condtons for a specfed perod of tme Tme untl falure of product or system. R(t) = P(T > t), MTTF, other statstcal defntons Instrument and measurement process nvarance (precson & accuracy, repeatablty & reproducblty) Consstency among ndependent measurement methods, ncludng temporal stablty, form equvalence, and nternal consstency. Kappa, etc Tme untl a non-complant (defect) event. Relablty = conformance rate. (how often get t rght, NCPPM)

8 Smplfed Defnton of Relablty Falure free performance over tme Focus: Tme between falures (longer s better) (Really a blendng of relablty, qualty, safety, and process capablty concepts.) Adaptng concepts from study of ndustry, safety, and hghly relable systems Provng to be a useful framework n healthcare 2004 Insttute for Healthcare Improvement

9 0 -X language 0 - : Sngle dgt falures per 0 opportuntes 0. to 0.9 falure rate (techncally.) Decmal locaton 0.x 0-2 : Sngle dgt falures per 00 opportuntes 0.0 to 0.09 falure rate (techncally.0) 0.0x 0-3 : Sngle dgt falures per 000 opportuntes 0.00 to falure rate (techncally.00) etc 0.00x 2004 Insttute for Healthcare Improvement

10 Relablty Levels Specfcaton - What relablty do you want? Same basc concepts and how termnology relates Relablty Falure Rate 0 -x PPM (parts per mllon) Sgma s , , , Insttute for Healthcare Improvement

11 Tme to falure n other ndustres Industry Arcraft engnes Sleeve bearngs Applcaton / Component Falure of turbne blade (crack exceedng ¼ nch length) Mechancal wear of bearng n a coolng fan due to frcton MTTF (hours of use) 0,000 hrs 30,000 hrs Computer chps Tme untl chp fals 5,500 hrs Power supply Power nverters 56,585 hrs Swtches Chargers 47,560 hrs 626,340 hrs Telecommuncatons Ethernet swtches,053,439 hrs Modems 55,424 hrs 2004 Insttute for Healthcare Improvement

12 Poor Relablty at all Sgma Levels Falure Rate 0 - Outcome or Process Measure EBM-CMS condtons (e.g., beta blockers for MI) Adverse events Hosptal readmssons 0-2 Deaths n rsky surgery Polypharmacy n the elderly Ventlator assocated pneumona (2-8 VAP/000 vent. days) 0-3 Adverse drug events (-6 ADE s / 000 doses) 0-4 Deaths n routne anesthesa 0-5 Deaths from major radotherapy machne falures 0-6 Deaths from sesmc noncomplance Qualty and Productvty Lab Northeastern Unversty, 2009

13 Relablty of Patent Safety Estmates (IOM 999, etc) Medcal errors & atrogenc njury: 98,000 deaths / year 770,000-2 mllon patent njures $7 - $29 bllon dollars Adverse drug events (ADE): 770,000 to 2 mllon per year $4.2 bllon annually Hosptal-acqured nfectons: 2-5 mllon NSI / year, $3,000 /case 8.7 mllon hosptal days 20,000 deaths / year More US deaths / year than for traffc accdents, breast cancer, & AIDS. Endemc AE s 6-0% hosptal patents suffer serous adverse events Adverse drug events (ADE) Surgcal ste nfectons (SSI) Needle stcks Wrong sde/ste surgery Devce-assocated nfectons - Ventlator-assocated pneumona - Catheter & central lne nfectons Per epsode average costs: ADE: $4,000 - $5,000 NSI: $2,000 - $3,000 VAP: 3 addtonal days & 30-50% attrbutable mortalty SSI: Can exceed $4, study: 95,000 deaths, $6 bllon/year Qualty and Productvty Lab Northeastern Unversty, 2009

14 Practce relablty Hysterectomy by age 70 Prostatectomy by age 85 Mane Hosptal Market A Mane Hosptal Market B Iowa Hosptal Market A Iowa Hosptal Market B Complance to Evdence Base Estmated only ~55% of U.S. patents receve scentfcally ndcated care: Acute: 54% Chronc: 56% Preventve: 55% Patents do not receve care n accordance wth best practces McGlynn, et al: The qualty of health care delvered to adults n the Unted States. NEJM 2003; 348: % 55% Patents receve care n accordance wth best practces Days n Hosptal Last 6 Months of Lfe Healthcare Systems Engneerng Northeastern Unversty Wennberg, 2005, From study of patents wth severe chronc llness who receved most of ther care n one of 77 best U.S. hosptals

15 Bundle example Many other examples Healthcare Systems Engneerng Insttute 8 Northeastern Unversty 202

16 Other bundle examples AMI CHF CAP CABG TJ VAP ASA (asprn) wthn 24 hrs BETA blocker wthn 24 hrs Thrombolytcs wthn 30 mn PCI wthn 20 mnutes ACE Inhbtor for LVSD Smokng counselng ASA at d/c BETA blocker at dscharge LVEF assessed ACE Inhbtor for LVSD Detaled dscharge nstructons Smokng cessaton counselng w w 2 w 3 w 4 O 2 assessed wthn 24 hours Blood culture pror to abx Frst dose of antbotc wthn 4 hours Influenza screenng / vaccnaton Pneumococ. vaccnaton Smokng cessaton counselng Frst dose antbotc hour pror to frst ncson Approprate antbotc selecton Dscontnued antbotcs wthn 24 hours Use nternal mammary artery (IMA) ASA at dscharge Frst dose antbotc hour pror to frst ncson Approprate antbotc selecton Dscontnued antbotcs wthn 24 hours Head of bed elevaton DVT prophylaxs Sedaton vacaton PUD prophylaxs 4 care elements 2. Center for Medcare & Medcad Servces (CMS), Clncal condtons and measures for reportng and ncentves Others: AHA, Leapfrog Group, Natonal Qualty Forum, AHQR, JCAHO Core Measures Qualty and Productvty Lab Northeastern Unversty, 2009

17 Relablty n dong what s known Recommended Care Receved ½ 64.7% Hypertenson 63.9% Congestve Heart Falure 53.9% Colorectal cancer 53.5% Asthma 45.4% Dabetes 39.0% Pneumona 22.8% Hp Fracture % Complance 70% 50% 30% Hand washng complance rate /08 2/08 0/09 02/09 03/09 04/09 05/09 06/09 07/09 08/09 09/09 0/09 /09 2/09 tme (Too Late) SSI Antbotc Tmng Example Average 6 Mns. Before St. Dev. 90 Mnutes (!) Optmal Wndow 0-2 Hrs Before Number of Hours Before st Cut (Too Early) Healthcare Systems Engneerng Northeastern Unversty 20 Not Optmal Patent non-complance Not No better: 36% 2 50% 3,4 Dscharge nstructons, exercse, meds, det, (eg, dabetes mgmt) ~ $50 B drug hosptalzatons ~ 40% nursng home admssons ~ $2,000/patent/yr offce vsts Total $77 bllon/year 2. McGlynn et al; 2. Chan DC et al ; 3. World Health Organzaton; 4. Natonal Councl on Patent Informaton and Educaton

18 2. why not better Healthcare Systems Engneerng Insttute 22 Northeastern Unversty 202

19 Desre vs. Desgn Wll and good ntentons (desre) alone nsuffcent Good well-meanng people are unrelable Human performance s fact of lfe Delberate desgn, that explots rather than gnores human factor Healthcare Systems Engneerng Insttute 23 Northeastern Unversty 202

20 Good ntentons, Bad desgns To Err IS Human Current System Forcng Functon Vsual remnder Redundancy Healthcare Systems Engneerng Insttute 25 Northeastern Unversty 202

21 Swss Cheese Model Nurse gves the patent a medcaton to whch he s allergc Medcaton Errors Fax system for orderng medcatons s broken Patent arrests and des Nurse borrows medcaton from another patent Tube system for obtanng medcatons s broken ICU nurse staffng J. Reason, 990 J. Reason and Johns Hopkns Medcne Healthcare Systems Engneerng Insttute 27 Northeastern Unversty 202

22 3. relablty by desgn Healthcare Systems Engneerng Insttute 28 Northeastern Unversty 202

23 Smlar deas Qualty Trlogy (Juran) Qualty control Detecton mtgaton (Some) Qualty Management as a System Qualty Improvement CQI, PDSA, etc (Most ) Qualty plannng Preventon ( st ) (Very lttle) Tend to be better at fxng thngs gone wrong than preventng n st place Relablty Model (IHI). Preventon strateges 2. Identfy, mtgate & mnmze mpact, recovery (QC, other methods) 3. Learn from falures, mprove system (root cause & falure modes analyss) Prevent Falures Identfy /mtgate that falure Redesgn to prevent future falures Qualty and Productvty Lab Northeastern Unversty, 2009

24 3-Ter Relablty Desgn Model Prevent Falures Identfy/mtgate that falure Mantan current level Redesgn to prevent future falures Improve relablty 2004 Insttute for Healthcare Improvement

25 All tems on protocol done 80% Every patent gettng lasx revewed by pharmacy for a dx of CHF CHF Example Prevent, Identfy, Mtgate CHF Protocol For All Admtted Patents 0% 0% Protocol Not Used Pharmacy starts the protocol f dx CHF Portons of protocol not used Best effort 0 - Best effort barely 0-2 Smokng advce (Global:All hosptal patents admtted automatcally referred to smokng cessaton f smoker) Portons of protocol not used (hghest falure modes) Detaled D/C nstructons (If protocol on chart clerk prnts out dc nstructon sheet at dscharge) ACEI use (If protocol on chart pharmacy checks for use of ACEI and calls MD f not ordered) Best effort 0-2 to a barely Insttute for Healthcare Improvement

26 Relablty Relablty Growth Over Tme 0-3 Prevent Identfy Prevent Identfy Redesgn 0-2 Redesgn 0 - Prevent Identfy Redesgn Tme 2004 Insttute for Healthcare Improvement

27 2004 Insttute for Healthcare Improvement

28 Relablty desgn/scence model Relablty ter Prevent Strateges Measures Process Outcome Detect Mtgate Redesgn Healthcare Systems Engneerng Insttute 38 Northeastern Unversty 202

29 Example: Patent falls preventon Relablty ter Prevent (survellance) Detect (assess rsk) Mtgate (mtgate rsk) Redesgn (refnement) Soluton Strateges Checklst Patent/famly engagement Process Use of tool on rounds Measures Outcome Percent complete / completed checklsts Assessment tool Usablty of tool? Use of tool Interventon assgnment per protocol to at-rsk patents (Falls Preventon Toolkt?) MySafeCare Gap analyss, RCA, and redesgn re: ncomplete check lsts etc Development and usablty of nterventon strateges # patent suggestons va MySafeCare Number of RCAs conducton Followng/ complance to nterventon process / protocol Number of process mprovements Healthcare Systems Engneerng Insttute 39 Northeastern Unversty 202

30 MGH CLABSI Example Relablty ter Prevent (survellance) Detect (assess rsk) Mtgate (mtgate rsk) Redesgn (refnement) Soluton Strateges Follow bundle by correct nserton, mantenance, access, and removal complance Audts on all process measures Infecton Control data Teamwork survey Educate new staff about CLABSI Performance updates (flp board, emals, etc.) Resdents dscuss lne removal on rounds Nurse and resdent orentaton addresses preventon technques Measures Process Outcome Reduce nfectons % complance to & resultng costs bundle Reduce mortaltes Survey data, # of audts done % of staff educated # of updates / month % of removals dscussed % of lnes taken out earler Identfy process shfts Improve care Number of process mprovements Healthcare Systems Engneerng Insttute Northeastern Unversty 202

31 Cumulatve desgn prncples Relablty desgn strateges Defect Level Strategy Prevent Identfy Redesgn Educaton Workng harder Feedback Checklsts, order sets Standardzaton Desred = default Opt-out vs opt-n Remnders Decson ads Redundancy Automatc checks Others? Some FMEA & RCA (but often ad hoc) 0-3 Forcng functons Less human relance Hardwred processes others? More formal FMEA Formal redesgn processes Healthcare Systems Engneerng Northeastern Unversty 20

32 Desgn Prncple : Preventon / 0 - Concepts 2004 Insttute for Healthcare Improvement

33 Basc Strateges: Intent, Vglance & Hard Work Common equpment, standard orders sheets Personal check lsts Workng harder next tme Feedback of nformaton on complance Awareness and tranng NECESSARY.but not SUFFICIENT 2004 Insttute for Healthcare Improvement

34 EXAMPLE: CHF Measures Left ventrcular functon (LVF) assessment Detaled dscharge nstructons ACE nhbtor for LVSD Smokng cessaton advce/counselng 2004 Insttute for Healthcare Improvement

35 Human Factors Strateges Decson ads and remnders bult nto the system Desred acton the default (based on evdence) Redundancy Schedulng Takes advantage of habts and patterns Standardzaton of process ESSENTIAL for movement to Insttute for Healthcare Improvement

36 Human Factors Concepts Vsual controls Remnders Check lsts Healthcare Systems Engneerng Insttute 48 Northeastern Unversty 202

37 Desgn Prncple 2: Identfcaton / 0-2 concepts 2004 Insttute for Healthcare Improvement

38 Redundancy Functon Develop a strategy to dentfy preventon defects Develop a strategy to mtgate the defects dentfed Develop a metrc to measure standardzaton falure 2004 Insttute for Healthcare Improvement

39 CHF Relablty Add Identfy and Mtgate 80% All tems on protocol done CHF Protocol for All Admtted Patents 0% 0% Best effort Protocol not Used Portons of protocol not used 0 - Every patent gettng major duretc revewed by case manager for a dx of CHF Case Manager puts the protocol on the chart Best effort barely Insttute for Healthcare Improvement

40 4. relablty strateges Healthcare Systems Engneerng Insttute 52 Northeastern Unversty 202

41 Smplfcaton and redundancy strateges Fewer steps Process smplfcaton (lean) Suppose each step 90% falure prob 2 3 n n Overall.90 90% % % % % % % More redundancy Suppose each step 40% falure prob n Overall % % % % % % % % % % Qualty and Productvty Lab Northeastern Unversty, 2009

42 Health care Industry Redundancy strategy examples Car brake system, arplane engnes, plots Computer back-up and RAID systems Extra brdge bolts and supports Multple readngs of lab results Double checkng prescrpton order Sponge counts Others.? Qualty and Productvty Lab Northeastern Unversty, 2009

43 Forcng Functon strateges Defnton: Desgn feature completely elmnates falure possblty Forces desred outcome to occur automatcally and absolutely Notes: Poke Yoke Toyota system Forcng scale some more absolute than others! Examples ATM card return Car door locks (some) Transmsson clutch Endoscope carryng case latches Soap dspensers to open doors Arplane occuped sgn (locks door, turns on lght)

44 Example One-sgma complance wth vacant sgn on bathroom door Current System Vsual remnder Redundancy Low level process desgn deas neffectve Compare to arplane forcng functon (lock, lghts) From Berwck Escape Fre 999 plenary presentaton, Commonwealth Fund webste

45 Systems example Leaky bucket example I can t pump desel nto my car, but staton can put desel n ther gas pumps! Remnders Tranng, Awareness Redundancy ❽ ❽ ❽ Human fatgue Poor lghtng at nght Sense of urgency Healthcare Systems Engneerng Insttute Northeastern Unversty 202

46 Desgn Prncple 3: Crtcal Falure Modes (Redesgn) 2004 Insttute for Healthcare Improvement

47 Crtcal Falure Mode Functon Develop a process to measure the most common falure modes Remodel the process based on the falure modes dentfed Iteratve process over tme 2004 Insttute for Healthcare Improvement

48 FMEA example Falure modes effects analyss RPN = Rsk prorty number = Severty x Occur x Detect Qualty and Productvty Lab Northeastern Unversty, 2009

49 5. actual relablty engneerng (and research) Healthcare Systems Engneerng Insttute 62 Northeastern Unversty 202

50 Healthcare relablty (engneerng)? Is ths really relablty (engneerng)? No, not really Blendng of qualty, safety, process capablty Ltmus tests: System stays n faled state untl repared or acted upon Can apply to just one tem Not ndependent dchotomous falures What would healthcare relablty engneerng look lke? Bath tub curve models Mean tme between falures (MTTF) Avalablty models Mantenance models Degradaton models Fault trees, block dagrams others

51 . Bundle relablty measurement (SPC) VAP measure Comply rate Head of bed elevaton 6% DVT prophylaxs 75% Sedaton protocol 43% PUD prophylaxs 88% Number Patents Composte Score Data Format # Satsfyng Measure Week X X 2 X 3 X 4 Incorrect lmts Correct lmts jb Composte Control Chart % % % % % etc etc etc etc etc etc etc etc etc Total Ops Week Total Met Composte Score

52 Northeastern Unversty Healthcare Systems Engneerng Insttute Exact probablty dstrbuton ) ( ) (... )... ( J J J w x w x t x J J w x w x t x k k w w x t X P x X P x X P J J J J k k k k Let: T W = w X + w 2 X w J X J and F W = T W /N, where X j ~ bn(n j,p j ) and w s unbounded Then: J W J W p p n w T V p w n T E 2 ) ( ) (, ) ( 2 2 ) ( ) (, ) ( J J W J J W n p p w F V n p w n F E J n T W p sw p (s) G ) ( t x w x t x w x w x t x k k w x w x t x W k k k k x X P x X P x X P x X P t T P )... ( ) ( ) ( ) ( ) ( Note: G T (s) G X (s) T not bnomal J J J J n p n N n p P n N P N and wth ),, ( Bnomal Naïve approxmatons 2 J J J J n p p n w n p n w ) ( and wth ),, ( Normal MLE) (MOM, ˆ,, m j j m j j n x p but ntractabl e,, ) ( ) ( 0 s T t t s G ds d t T P P(F = f) = P(T = t), where f = t/n, Means, Varances, Probablty generatng functons Probablty dstrbuton (convoluton) Naïve approxmatons? 2 3

53 2. Optmal amount of redundancy Optmal amount of re-nspecton n reuseable medcal equpment to prevent cross-contamnaton (n press) Governng equatons: Qualty and Productvty Lab Northeastern Unversty, 2009

54 k of n (and falover) relablty models Bundle effect If at least k of n measures are met for a patent, the lkelhood of adverse outcome sgnfcantly decreases Non-lnear bundle effect.0 Concept Illustraton 2 k n Probablty No AE 0 "tppng pont" 0 k n

55 3. Relablty bathtub curves Early falures < ) Random or common cause falures ( = ) End-of-lfe falures ( > ) In Theory Decreasng hazard: Recovery from a surgery Increasng hazard: Health Deteroraton due to old age Hazard rate Decreasng hazard functon Constant hazard functon Increasng hazard functon Constant hazard: Accdent or rare dsease Tme Falure Modes: Strateges: Producton falures, nfant mortalty Burn n, screenng Envronmental falures Useful lfe, survellance Degradaton falures Warranty, obsolescence Hazard functon of entre human lfe cycle f ( t) d h( t) P( t T t dt) log R( t) R( t) dt Dfferent causes and strateges for handlng each phase Dfferent shapes for dfferent processes Mortalty Potental Applcatons Heart falure and other chronc llnesses Hosptal readmssons Returns to OR Cancer recurrence

56 Mortalty hazard functon Mortalty hazards by year of brth: 900, 925, 950, 975 (Top reasons for death by age also are lsted n rank order, 900) Mortalty Rate 20th Century Mortalty Rate/00, Influenza & neumona Tuberculoss Cardovascualar dsease Dsease of heart Accdents Cerebral dsease Accdents Tuberculoss Cardovascular dsease Cardovascular dsease Dsease of heart Neoplasm Cardovascular dsease Dsease of heart Neoplasm Cerebral dsease Dabetes Chronc pulmonary dsease Pneumona Dsease of heart Cerebral dsease Accdent Under year-4 years5-4 years5-24 years25-34 years35-44 years45-54 years55-64 years65-74 years75-84 years85&above Age ^ Natonal center for health statstcs data bank

57 Root cause analyss Top Reasons for Infant Mortalty Influenza & pneumona Influenza & Pneumona Prenatal condtons Tuberculoss Tuberculoss Congental anomales Cardovascular dsease Dsease of heart Accdents Cardovascular system Pneumona & nfluenza Accdents and adverse effects Brth njury and labor dffcultes Congental anomales Symptoms and ll defned condtons Pneumona and nfluenza Accdents Dsease of heart Dsease of heart Accdents Cerebral dsease Cerebero vascular dsease Brth defects Pre-term or low brth weght Sudden nfant death syndrome Pregnancy complcatons Respratory dstress syndrome Labor dffcultes Infectons Accdents Pneumona or nfluenza ^ Natonal center for health statstcs data bank

58 US and UK male death rates, 997, for whole country (n and out of hosptal) vs agegroup death rate/00, < '-4 ' agegroup UK US US whte US black

59 Dsease hazard functons Cancer Mortalty 200 Heart Falure Mortalty Rate Rate of Death Mortalty < -4 yrs 5-9 yrs 0-4 yrs Under to 4 5to Age Age Hazard rate for chronc respratory llness Intestne transplant falure h(t) < -4 years 5-4 years 5-24 years years years years years years years > 85 5-Jan 0-Jun 7-Nov Tme

60 Process hazard functons 70% 60% Early Falure Root Causes 50% 40% 30% 20% 0% 0% Poor Dscharge Surgcal Complcatons Poor Hosptal Procedure/Rx care not successful Infecton Other 70% Chance Falure Root Causes 60% 50% 40% 30% 20% 0% h(t) 0% Poor Outpatent Management Lack of Support LTC Faclty Problems Non Complant Unable to get Unable to get Patent meds appontment Days 2 3 Data follow left-hand sde of bathtub curve Decreasng falure (readmt) rate as patent tme from dscharge ncreases

61 Falure modes, dfferng over tme What to focus on when 70% Early Falure Root Causes 60% 50% 40% 30% 20% 0% 0% Poor Dscharge Surgcal Complcatons Poor Hosptal care Procedure/Rx not successful Infecton Other 70% Chance Falure Root Causes 60% 50% 40% 30% 20% 0% 0% Poor Outpatent Management Lack of Support LTC Faclty Problems Non Complant Patent Unable to get meds Unable to get appontment

62 4. Avalablty and mantenance models Tme tll falure (l 2 ) Operatonal State p Repar tme (l 2 ) Faled State p 2 Human health Patent = complex system Falure = Degeneraton from healthy to sck state Repar tme = Treatment and cure tme Avalablty = Proporton tme n healthy state Emergency Dept. Dverson ED = complex system Falure = ED deterorates to pont of closng Repar tme = Tme to get off of dverson Avalablty = Proporton tme ED open to ambulances

63 6. Human factors engneerng Mental load & human performance (queung network model) Drvng/plot smulator (NASA) Healthcare Systems Engneerng Insttute 85 Northeastern Unversty 202

64 Done!