《research in statistical quality control,.pps》由会员分享,可在线阅读,更多相关《research in statistical quality control,.pps(8页珍藏版)》请在得力文库 - 分享文档赚钱的网站上搜索。
1、James C.Benneyan,Ph.DDirector,Quality and Productivity LaboratoryNortheastern University,Boston MA 02115www.coe.neu.edu/Research/QPLbenneyancoe.neu.eduAbout QPLThe Quality and Productivity Laboratory conducts theoretic and applied research in statistical quality control,experimental design,process i
2、mprovement,and applied probability in a variety of industrial and service applications.Recent work focuses on statistical characterization of powdered metal production processes,healthcare quality and patient safety,optimal cancer screening policies,statistical methods for rare events,experimental d
3、esign in the printed circuit board industry,inspection error and multiple inspection cost models,and approaches to multi-response optimization.The lab maintains collaborates with several associates in academia and industry,with funding from the National Science Foundation,the Agency for Healthcare R
4、esearch and Quality,NIH,and industry partners.Various QPL members hold joint appointments,Board positions,associations or editorial positions with the Institute of Industrial Engineers,Society for Health Systems,Raytheon Six Sigma Institute,Institute for Healthcare Improvement,and numerous professio
5、nal journals.Current Focus AreasRare event quality control methodsRisk-adjustment statistical methodsBest practice identification(data envelopment analysis)Self-adjusting quality control chartsSafety,reliability,experimental designOptimal cancer and lab screening modelsFundingNational Science Founda
6、tion(NSF)Agency for Healthcare Research and Quality(AHCRQ)Industry partnersQuality&Productivity LabQPLRisk-Adjusted Statistical Quality Control MethodsBackgroundNon-identical binomial populations.Manufacturing:Production lines,Plants,WidgetsHealth Care:Hospitals,Departments,Procedures,PatientsStanda
7、rd control charts are incorrect.Risk-adjustment SPC framework:n0=10 p0=0.99n1=100 p1=0.01ResultsExact j-binomial probability distribution derived and implemented numerically Several new exact Shewhart,EWMA,and CUSUM methods derivedMathematical proof j-binomial variance 0 for rate increases;for rate
8、decreases n=1(g chart)and n=5/p(p chart).Examples are shown below.Next StepsComplete coding of CUSUM portion of program and conduct more thorough analysis and comparison of the various methodsQuality&Productivity LabQPLBackgroundNew control charts based on geometric and negative binomial“inverse sam
9、pling”for monitoring Bernoulli processes,especially those with low defect ratesDeveloped numeric code to calculate average number inspected(ANI)for g and p charts recursively.Implemented search algorithm to identify optimal subgroup size for each chartPerformance investigated for k-sigma and probabi
10、lity-based one-sided and two sided control limits ResultsOptimized one-sided g charts always outperform p charts for all nonconforming ratesOptimized two-sided g-charts have significantly lower ANI to detect increases(verify improvements);two-sided p-charts have have lower ANI to detect decreasesFre
11、quently advantageous to use n 1 in design of“number between”g charts,contrary to convention in literature3 journal papers currently in preparation describing resultsExample of ResultsComparison of Optimized g and p ChartsOptimal Geometric and Negative Binomial Statistical Quality Control MethodsQual
12、ity&Productivity LabQPLFocusMethods for controlling healthcare processes with some type of stationary or non-stationary autocorrelation structure,including:Combined SPC/bounded feedback adjustment schemes(with deadbands),such as to maintain insulin levels in desired range for diabetic patientsMethod
13、s for monitoring autocorrelated healthcare processes,such as seasonal respiratory illness ratesBackgroundRecent literature interest on integrated SPC/EPC feedback schemesLittle-to-no work on Box-type bounded(deadband)adjustment schemesParticularly relevant to healthcare and other industries with cos
14、ts and consequences of continual adjustmentPreliminary ResultsBounded Adjustment Scheme Applied to ARIMA(0,1,1)Disturbance ProcessNext StepsApplication of bounded/unbounded feedback schemes and autocorrelation SPC methods to healthcare processes(in conjunction with IHI,MGH,and BCH)Research on altern
15、ate SPC methods to monitor bounded adjustments in order to detect changes to underlying disturbance model&/or autocorrelation structureDevelopment of cost optimization models and sampling interval modelsInvestigation of method robustness to model assumptions and mis-specificationOptimal Integrated B
16、ounded Feedback Adjustment SPC Schemes for Patient Monitoring and ControlWithout any process shift With trend of magnitude 1 per unit introduced after 55.data point Quality&Productivity LabQPLA.Multivariate Loss FunctionsBackground:Quadratic loss functions have been used to find the best balance amo
17、ng process control variables and can be extended to minimize total expected loss of multiple responses.We combine this idea with minimization of the total loss variance by deriving the theoretical loss variance and probability distributions for several response distributions.These results are used i
18、n a multi-criteria optimization model which minimizes total expected loss subject to a minimal valued variance loss and a risk constraint that keeps the probability of the total loss above a specified level.B.Frontier Estimation ProblemsBackground:Data envelopment analysis is a non-parametric fronti
19、er estimation method that constructs production possibility sets from empirical inputs-outputs of multiple decision making units(DMUs).In practice,missing data makes application of standard methods impossible,although little research addresses solutions.Simple proposals include removing missing data
20、 from the analysis or replacing missing outputs(inputs)with a 0(big M number);neither yields acceptable results.Possible alternatives involve statistical methods,multiple imputation,and least squares estimation.A related problem of practical importance is the need for methods to partition large numb
21、ers of DMUs(e.g.,all U.S.hospitals)into peer strata,since knowledge of as many as 7,000 DMU scores alone is not always that useful in healthcare improvement projects.ExampleSome Optimization Problems in Multi-Response Quality ControlQuality&Productivity LabQPLBackgroundSeveral programs and software
22、tools have been developed to facilitate the labs research and to enable practitioners to use the developed methodsTools section of website expanded,all tools to be made available over web-site(w.i.p.)Summary of Programs1.Implemented new statistical quality control methods in user-friendly GUI-based
23、softwareAll standard methodsNew Shewhart,EWMA g,h,j-binomial and standardized methodsCUSUM and probability-limits yet to-be added2.Java-based g chart,performance,and optimization program3.g and p chart performance and sample size optimization research-level code4.MC programs to compare EWMA and CUSU
24、M g and p charts5.Automated spreadsheet templates for g and j-binomial convolution methodsSpecial-Purpose Software Development and Website ToolsJ Binomial EWMA Standardized ChartEWMA g Chart with 3 Sigma LimitsQuality&Productivity LabQPLBenneyan JC,“Performance of Number-Between g-type Statistical C
25、ontrol Charts for Monitoring Adverse Events”,Health Care Management Science,2001;4:319-336.Benneyan JC,“Number-Between g-type Statistical Control Charts for Monitoring Adverse Events”,Health Care Management Science,2001;4:305-318.Brown SA,Benneyan JC,Theobald DA,Sands K,Hahn MT,Potter-Bynoe GA,Stell
26、ing J,OBrien TJ,Goldmann DA,“Use of Moving Averages and Binary Cumulative Sums in Nosocomial Cluster Detection”,Emerging Infectious Diseases,2002;8(12):1426-1432.Benneyan JC,Borgman A,“Risk-Adjusted Sequential Probability Ratio Tests and Longitudinal Surveillance Methods”,International Journal for Q
27、uality in Health Care,2003;15(1),in press.Brown SA,Benneyan JC,Theobald DA,Sands K,Hahn MT,Potter-Bynoe GA,Stelling J,OBrien TJ,Goldmann DA,“Automated Detection of Nosocomial Outbreaks-A Comparison of Two New Alternate Systems”,2002;under review with Infection Control and Hospital Epidemiology.Benne
28、yan JC,Borgman A,“The Exact Probability Distribution of Non-Homogeneous Dichotomous Events”,2002;in preparation for Journal of the American Statistical Association.Aksezer C,Benneyan JC,“Variances and Probability Distributions of Quadratic Loss Functions”,2002;in preparation for European Journal of
29、Operational Research.Benneyan JC,Nene AM,“The Optimal Subgroup Size for Number-Between Control Charts”,2002;in preparation for Journal of Quality Technology.Curran E,Benneyan JC,Hood J,“Controlling Methicillin-Resistant Staphylococcus Aureus:A Feedback Approach Using Annotated Statistical Process Co
30、ntrol Charts”,Infection Control and Hospital Epidemiology,2001;23(1):13-18.Benneyan JC,Lloyd RC,Plsek P,“Statistical Process Control as a tool for Research and Health Care Improvement”,to appear June 2003,Quality and Safety in Healthcare.Benneyan JC,Design,Use,and Performance of Statistical Process Control Charts for Clinical Process Improvement”,2002;under review with Quality and Safety in Healthcare.Publications and Dissemination