Guest editorial
‘The Sith Stu Hunter Research Conference was held
March 6 ~ 8 2018 at the Roanoke Hotel in Roanoke,
Visginia. The Hunter Conference strives to provide
Thoroagh calls fllowed by a robust discussion among
savance the research fla. Daring the conference, sit
Ieynote speakers were vena forum of 90 minutes to
present thelr ideas, after which two dscusants chal
Tenged the ideas, and provided an eacellont basis for
the pen discussion with al participants. The lasses
tion ofthe conference was devoted to the late Soren
Bisgard who had a tremendous impact om the Held
of industrial satis,
‘his special Isue provides fll-ength papers of all
six Keynote speakers Fach ofthese papers is comple
rented by st lee wo discussion papers that are pro
vided by the dacussnts atthe conference, x8 well
by some addtional authors wna are wel respected in
the fed. The speaker. thelr topics, and the invited
dlscuesnts included the following
‘Dennis Lin, Pens State, “Design and Analysis for
Order of- Addition Experiment”
Discussant 1: Joo Voc, Rocheer Insiae of
Tecnology
Disusent 2: David Stenberg Tel Avr Unery
aura Freeman Institue for Defoe Analysis
“Design of Experiments for Refit”
Diecast
Dicusent 2 Zhisheng Ye, National Universty
af Sgepre
Sandy Fogel, Vigna Tech Cariton Schoo! of
Medicine, "Doctors ate ot pilots and patents ae not
seplanes: Quality improvement in medicine”
Discussant: Stan tinct, Unies af Waero
Doug Montgomery, Azone
Discussant 2 Vitra Jordan, Emory Heal Care
Stephanie Kovalchik, Tennis Australia, “Gaining an
Bdge Promoting Stattial Thinking in the
poms Industry”
Taylor & Francis
aoe
Dicusant 1
State Uniersiy
Kan Pander, Nor Guolina
Discus 2: Steve Rigdon Saint Lous Univers
Ross Sparks, CSIRO, “Disease surellnce: Disease
‘Sa Franco.
James D, Wison, Univesiy. of
hacen 2: Fade Mabe Vigna Tech
Geoff Vining, Veginia Tech, “Soren Bisgard's
Contributions to Quality Engineering”
Denmark
Murs Kulhcl, Tena Uniersty
Diacsant 2 Roger How Union Coleg
There are additional discussions in this Bsue of
Quality Enginering from participants of the 2018
Hunter Conference who were not invited discussant
‘We thank Murat Caner Test, past editor of
Quality Engineering, Marcus Perry the curent editor,
and the staf at Taylor and Francs for allowing us the
‘opportunity to serve as guest editors and to present
5 papers in this issue
‘We are abo gratefil to the Hunter Conference
‘managing commit which consists of Stefan Steiner
and Musat Kulahel, and the local commits which
incuded Aane Driscol, Geoff Vining, Row Fricke,
Kort Lindssy, and Jeneen Preston. Finally, we are
iratefil to all sponsors of the conference, including
the Stasis Division of ASQ, the Chemical and
Process Industries Division of ASQ, the Quality and
Produesvty section of ASA, and JM
Bart De Ketuere
Steven E. Rigcon
Sunt Lous Unies
‘Changang Zhou
Nate Univerty
T
Doctors are not pilots and patients are not airplanes: Qual
in medicine
Sandy Lewis Fogel
Taylor & Francis
improvement
Daren Supe ig Tech iin Schou of Mei, ok, Vega
tied Ta Er i Human” Wt descbed enor howls which te Yo between 44000 and
‘S000 dente per year Thi wae the “tpping pote’ tonardslfevsualon and quality
orrecay
soy
itmpronment Nesine nar slowly deveaped Quy iprovenent methocloges tat
iimrto work Suge lea the wy. Tie ae ll go trough same the metry ot
“ay improvement th the caty comparing poets ond celng eutore.
Steh toe Stele chang phyn teravor The challenges of mplmentng ty
Tipronent scoss anette Nasa ibe demonstrate
History
For much ofthe history of onganied medicine in the
United States, quality improvement was in the form
of peer seve. This took place only when something
tunexpected and sawanted occured to a patient. It
sees felt that the physician In charge was solely
responsible. A group ofthat physician's pers gathered
ina room and discussed the negative outcome. If it
wes the consensus ofthe group that something wat
done incorrectly, the phyician who made the exzor
fo the proverbial slap on the wrist. That physician
‘would theoretically never make that mistake agai:
however, this proces dd nothing to prevent the next
physic fom making the same error. Nor dd it pre
‘ent the fst physician from cresting injury 10 othar
Patients ina similar but sot identical manner. This
approach did nothing to address systematic erors of
to provide gystematicslutons
‘The Ais Known attempt to se dite to eee sy
tematic quality improvement took place in New York
State in the late 1980s, The sate plished the mortal
ity rte ofall howptal that performed cardiac surgery
This was intendd to allow the public to know the hos
pitls with the lowest morality and seek caveat those
facies. Thi increased paent choice was intended to
multe the other hoeptale to improve in order 10
‘compete, Calbia Univers in New York City atthe
time was generally considered o be one ofthe premier
‘nsttutons for cardiac sargey. I wat the efertal een
ter for many other hospital in the cy, 8 well a in
the state: It was where tbe most complicted patients
were se. A sures finding inthe publication of mor
tality data was that Columbia had one of the highest
mortality rates in the ste, Thi engendered 2 furor of
press conferences and letters to the editor intended t0
defend Columbia. When calmer heads prevaled, it
‘became obvious that of course, Columbia would have a
high moray rate; they saw the sickest patient, many
of wom had been turned down by other isiutions
for being too high risk What the publication of the
data actually accomplished was creating an inceive
for hospital to tan down high-siee patients and only
fperte upon the lowest rok patients inorder to ep
thelr mortaity rate down. Alhowgh the term bad got
yet been used in medicine, what vas mended was some
form of rik adjustment to the dat.
“The next development grew out ofthe aftermath of
the Viemam War which ended in 1975 Returning sol
ders often suffered the brunt of the anger of the
“American populsee. By the middle of the 1980s, the
[US Senate began lange about ow to help the vet
rans. Complaints began to surface about the level of
care cflered by the Veterans Adminstration (VA)
hovpitals The perception was that shoddy surpcal
Taree pound aS Sette mae
a a a ng ee
‘Sewn afm etme the tes ne ren be sd wwe.+ @ sure,
care was beng offered, In 1989, the Congres: pase
the US public aw 98-166 which, among its many
ges da clause dat stated thet the quality of sur
peal care a the VA hospitals will be Brought up to
national standards within Syears. Nobody would Took
tt such @ statement and believe that was not lofty
snd admiable goa
"The callenge of bringing this change to fation
was taken up by « group le by De shukri Khari 2
Professor of Surgery at Harvard well as a PRD. in
Epidemilogy. One of the Bret things this group rel-
taed was hat there were no national sandards. They
didnot know how eo raise the standards at the VA
Hospital m an unknown bar. They also realized that
they bad no way t0 meaice qualty. Without tis
measure how could they determine if care had
limprovee? They dlacovered that there was no way t0
compare rospitul, Hoeptals that serve « needy and
Lunderserv patient population with Ile to no access
to primary care cannot be compared to hospital in
tan upperlse saburban neighborhood where all the
Patients ae Insure, have good primary care, are well
hourished exercise rgulay, ‘ake their medications
felgiouly and are overal alte healthy. Finally. they
realized tat there was no deflation of a quality surg-
fl oatcone. The measurement of moctaly $62 705s
Slapliicrion of the sue, A surgery can have 220
‘mortality but bave a huge mumber of complications,
Jong hospital say and unhappy patients, For te su
fgeon, a sucessful outcome may be diferent than
what the patent views as a successful outcome,
‘By 196, all 138 hospitals ln the VA gystern were
entlled‘@ the progr, and Dr. Khu had overall
fveright. The program installed the first nied dec
tronle madical record which linked all 133 hospitals
This was in essence, an Intemet where Dr. Khuri
could have acces 10 the medical records of ny VA
hospital while siting in his office in Boston
Hundreds of variables were collected on each patient,
and an ongoing statistical analysis was performed. The
{group erentually found 62 preoperative arabes
(Gach as patient age weight, and preexisting medial
condition) and 30 intraoperative variables (uch as
Aluration of operation, blood loss, and patient tem:
perature that were the independent variable, affected
41 portoperatve outcome variables (such as death,
Infect and other complication) as the dependent
variables Stepwise logistic regression was performed
to determine the maghitude of impact of each af the
Independent variables upon the dependent varlables,
land a nuke model was generated, This as the frst
secious attempt at sak adjustment. The outcome data
were presented ae observed to expected (OIE) ratios.
‘Nurses at each participating hospital were brought
to a central location and trained on the definitions of
cach variable and on the accurate collection and
‘uploading ofthe data Because ofthis rigorous proces
Of sanderdization of definitions and trsning of data
atstracters, by the Year 2000, the measurements had
become reliable and reproducible over ume, and the
measurements had become rlable and reproducible
between oss, Hospitals could be reliably and
reproducibly. compared within the dskadjused
‘mode. All 133 hospitals in the VA sytem improved
signieatiy. Over the entre VA system, there was 2
2% fallin mortality and a 45% fll in the total com:
plleston rate (Kharl et al. 1998)
‘The Institute of Medicine report
[At sbout the sme time that De. Khus was observing
resus In his VA projet, The Institute of Medicine
(GoM) of the National Academy of Sciences produced
f report in 1999 ened "To Ere ie Human" (Kohn,
Corrigan, and Donaldson 2000). I stated tht between
“44000 and 99,00 patents die in hospitals each year
from preventable errors. twas a ples to Dall a safer
health system. The intial reponse was a predictable
voviferous denial along with attacks on the method
logy tied by the TOM to determine these gues,
(Over a short period of time, however, this became
ccepled a6 likly true and possbly the tip. of
the iceberg.
‘Shorlytherefer, « number of books designed to
suldess the [oMs ell for improved safety in medicine
‘were published. One such book was ented Practicing
‘Meine in the 2st Century (Nash eal 2008) twas
1 description of how computers and other technology
ould replace mach of physicn's work with algo
rithms that would create far fewer ercors and lead to
far fower deaths, This book, along with much other
erasure, had a profound effect on the US federal
goverment which eventually mandated elecironke
Iedcal reconds (EMRs) in hospitals and in phys
ican’ offces, Technology ceralaly prevents some
errors but, unfortunatly, creates others. The final
trord as fo whether or not EMRS have made bealth-
Cre safer il out, but they cetaialy do pall pro
vers avay fom the bedside (Higgins et al. 2017)
“Another book publshed net the sime time was ent
fled Overeated (Brownlee 2007). The author was a
reporter who showed many examples of “why 00
much medicine is making us sicker and poor” It was
4 plea fo simply do less. This didnot x the problem.
“Another example of » book in this vein was weiten
by a hospital exectve whe left medicine and joined
the Disney organization and war ented If Disney
‘ax Your Hosptal (Lee 2004) This fostered a move
within hospitals to address the “hotel” functions. It
snas felt that if the hospitals lmproved their decor
food, linen service, tc, enors would diminish. They
id et
“There wes then a book writen by an atlorney ent-
tied Why Hospitals Should Fly (Nance 2008). 1
described the vast Improvements made in the ailing
industry inthe 19708 by going through a series of
checklite and eliminasng the calture of deference
“This worked tremendously well inthe adine industry,
and it was felt a if this could also work in medicine
While yng a plane, when the pilot tums the wheel
to the sight, the plane goes to the sight. When the
pilot pushes the landing gear lever, the wheels go
down. The response ls predictable and reproduce
(Quality in manufacturing may involve creating just
sach 4 consistency of expectations. Patients do not
respond that way. Some patints start out quite
hralthy and have predictable ovtcomes, but othees
Ihave complex preexisting conditions such that the
response to an intervention seems random. For
ccrample, patents may be malnourished or immune
‘compromised, both of which increae the expected
rortalty and complication rates Obesity, diabetes,
heart discase, and numerous others of the preopers:
tive rik factors alter patent respons to teament
[Additonaly, ao two patients have the same genetics
‘A medication that work for one patient may be toxic
to another. To complicate testment rests further,
‘many ptients are not compliant withthe tcatment
plan, One begins to get the ida that every patient is
Ubierent, and we are not smart enough to take all of
the billions of combinations of genetics, preoperative
Fath, and patent withes into account to predict oxt
comes, Some improvements were found by becoming
Imore rigorous with checks and empowering anj-
body who cares fora patient to speak up with quality
conceans, especialy in preventing the “never cvets”
Sich as operating on the wrong limb oF the wrong
Patient, However, the impact on overall quality was
Elnply too small to be the only solution (Treedwel,
seas, and Toa 2014)
Nsaw
‘The American College of Surgeons (ACS) ie the
umbrella organization of surpeal societies in this
aur aoneane 5
‘country, The vast majority of surgeons belong, to the
ACS, and iti the source of much of the information
‘we have about overall changes in surgery. In 200, the
[ACS bought the rights tothe VA. quality improvement
process called the Natonal Surgical Quality
Improvement Program (NSQIP), A stries of alpha
land then beta hospital sites tested the program and
oved that it worked 3s well outside the VA (Hall
f@ al 2009), By 2005, Individual hospitals were
encouraged to join the program and have designated
nurse reviewers trained. This was fait expensive, and
the proces wat me-consuming making inital op:
tion of thi program relatively slow. But overtime
more and more hospitals began to join the program
making the database larger and more robust. There
fre currenly over 750 hospital enrolled covering the
ast majority of all surgery in the country. Ris now
fone of jst & few validated databases in medicine
Stepwise logistic regression is used to crete risk mod:
fs. with results initally presented ax observed to
expected ratios and later as odds ratio. This isa valid
fan robust metnod of ri adjustment tha allows hos:
Dials to be compared and to work toward improve
nents Although it isthe best we have is sil fr
from perfect. Stepwise logistic regression will under-
‘estimate the interaction of varbles, an important
fonsderaton in a system as complex as ¢ human
beng (Livingston 2010). All 62 preoperative variables
and 20 Inteaopertive variables are considered Inde-
pendent in the stepwise logistic regression, but they
fre not. The patient vith severe fection (sepsis) is
flea likely to have kidney fare, cespictory fllre,
fand other associated symptoms or conditions. Many
ofthe variables are linked. Livingston also stated that
the seprise logistic reresion process will overest
mate the significance of sporadic associations that will
‘cca bese ofthe large numberof variables,
Tn 2010, NSQIP changed the procedure risk sore
(Rael etal 2010) This is @ factor nthe equation
that takes into consideration the diferent risks inber-
ct in differen surgical operations. In 2013, there wat
4 change (0 hierarchical modeling. with shrinkage
tdjustnent for small sample staes (Cohen et al 2013)
In 2015, i was sated tht Hf a sgnfcant variable that
affects the outcome Is left ot ofthe model there Isa
Jarge eet on the performance ofthe mode (Etzioni
al 2015), Such a variable might be preoperative
fnarctle dependence. ‘There sigaliant lterature
stating. that this ads rk and. worsens outcomes
(Apfel et al 2012; Lee et al 2014), but it is not
Included in the NSQUP vik model In 2016, it was6 @ street,
reported that diferent modes were needed fr emer:
gency versus cetve procedures (Hyder et al. 2016),
The comparisons between hospitals were fest pee-
sened as caterpillar plots and then were changed to
being reported by docks, Woodall subsequent stated
that «beter way to present data to foow an individ
ual physician for Binary outcomes, such asthe pes
cence or absence ofa complication, would be through
2 Bernoull-ajasted RA-CUSUM chart in assocition
withthe use of the VLAD plot (Woodall, Foge), and
Steiner 2015) And it has been suggested that a better
vray to compare inshutions ie withthe use ofa fannel
lot (Spiegelalter 2005) Cleary, the optimal method
to develop risk model and to present dat in surgery
1s a moving target.
Quality improvement
Soy why is quality improvement so hard? Do not all
physicians want to do the right things for their
patents? The iajor cesson ie habit. Physicians get
Fnto the habit of doing things the came way every
time, Surgeons waat t0 do an operation exactly the
same way, with exacly the same instruments, with the
steps in exactly the sme sequence. Each surgeon all
bvotely master the processes Being esed fo a par-
tlclar type of opertion. That surgeon wil get reliable
and predictable resus. The fact that they may not be
the Dost rele is lmmaterial. Ifyou ale che surgeon
to change all of «sudden that surgeon isnot the mas
ter of what they’ do, A surgeon may have done a par
ticular procedure $00 tates with a 2% complication
rate. A new procedure may be developed that leads to
21% complication rate, Unfornatly, the fist few
times a surgeon docs new procedure their inxper
ence with the operation may cause the complication
fate to be considerably higher than the expected 1%
‘Would you want your surgeon to do it the ew way
forthe way they have done i 00 times?
“The nes ese i insttional enoul An institution
develops «cure that is passed down in a consistent
and predictable fashion. Change in the proceses of
patent care is not only changing physician behavior
but also itis changing the behavior of everybody that
comes in contact with the patient. The dials of
‘eueating multiple physician growps, nurses, nurse's
fides, cespirstory therapists, physical therapists, et,
fare extraordinary. Hospitals work 24hours a day,
7 days a weck, and 365 days a year. Thee it n0 down:
time to provide employees additonal tsning. ‘To
change processes frequenly means having. people
ome in on off hours or pling people from thee
patient cate responsibilities for waiing. That means
having addtional people avalaleto replace those not
feviag patient care. And this lads tothe third major
eason why it is hard to improve quality in medicine
‘money, Hospitals work on rgorthin margins and
ny addtional money that bs spent needs to have a
return on investment It is hard to document such a
return forthe prevention of complications that don’
take place in the future, These three retons deve the
mated I7yeur between the time something is
documented in the Ierature as the best process for
fare fora particalar problem and is adoption within
‘the medical establishment (Brownson ea. 200).
Because of all of the above challenges, quality
Improvement in medicine is quite hard. However the
processes requlted to Implement quality improvement
‘changes are being Incressingy incorporated by hosp-
tals across the country, There are muliple methods to
work toward quality improvement such as lean, 6
Sigma, DMAIC, and others. In one way of thinking,
they are all simular, and the proceas comes down to
five principal steps. The Bs 6 defining and collecting
data Thre isa quote often atibuted to Drocke that
fates “Wf you cannot measure, you cannot improve
Ia” Although thin is probably miscuote, 1 belive
this is true in medicine. Outcomes are so subjective
that if we cannot put a number on improvement
fffors ae jart gueawark Another quote, this one
sterbuted to Deming sats "In God we trust All
thers must being dita” Physicians are not very good
ft data. Moet physicians do not know thee own data,
fd do not want to know thir awn data: When pre
sented with data related to thelr performance, phys
‘ans become very defensive and attack the valid of
the data. But with continued sxe and presentation of
‘the data, most physicians eventually Bega to tas
data Beng shown,
The second step in the process Is the analysis of
the data, Without knowing what the data mean, the
data are jase arbitrary numbers. A surgeon who Is
told thelr surgical site infection rate Is 25% has a
‘umber in iolation. Is that good or bad? Compared
{o what? What i the benchmark? Physicians ae actu:
ally quite good at analysing the dat. Once presented
wth’ data st, and after the Jniual denial phase,
Dhysicians understand and interpre the consequences
ofthe contexvalied dat quite wel,
The third step is making plans for improvement
based upon the dats, Physics are moderately good
at doing this. They are very good a telling other peo-
ple what they need to do to get better, but not Very
food looking at themecver inthe miler and
Table 1 A pre st of tose inva sual quality
i =
=.
— ome
ie
rests 1 enay 207 t 31 Deeb 2007.
‘on clon iso
tile te 8 ay pnt is
ee eo
2 3
ee 3
Aetermining what they personally need to do 10 get
better However, with continued presentation of the
data this ls becoming esential and accepted within
the physician community.
“The fourth step in 2 quality improvement process
fs implementation of the plans. This is by far the
ost difical step, This is where many hospital
This takes a fot of time, a lot of phone calls lot of
emis, a lot of meetings, and # champion for the
‘quality improvernent process that ls driven and deter
‘ined to make thie happen. This is where cooperation
between the varying profesional aswell a6 ancilary
staf In the hoeptal is partmount. This obviously
Includes physiins and nurses but also inciades phar
‘rains, detdans, and others to numerous to count,
«part Ist of which Is presented in Table 1.
The final step i obtaining new data It needs to be
etermined s€ the quality improvernent project has
scwally improved the quali. Has the bar been
moved? If i has not, then there i eltber something
‘wrong with the implementation, something wrong
conmrananens 7
wth plans, something wrong with the analy, oF
fomething wrong wih the dats, The team working on
the quality improvement project needs to work back
‘ward and find ut where the problem lis. This needs
to be adjusted, and the ve steps followed once again
to assure tha the bar indeed gets moved
‘Thee is another quote attributed to Deming which
rads “every system is perfectly designed to get the
results it gets” This meins that you cannot just wor
harder within the system to get better outcomes, The
system has tobe changed. The experts at NSQIP have