Lauren Silver HPM 582: Design Analysis #1

1. Obesity, inactivity, and the prevalence of diabetes and diabetes-related cardiovascular comorbidities
in the U.S., 2000-2002
Objectives: Both a sedentary lifestyle and being overweight and obese often occur simultaneously in the same
individual and, thus, are associated with higher levels of diabetes and related cardiovascular comorbidities.
Given this connection, this study sought to assess the independent relationship between physical inactivity and
body mass index (BMI) with diabetes and diabetesrelated cardiovascular comorbidities. In other words, is
adiposity any more important for health than fitness level or vice versa!
Design and data sources: "his study utili#ed observational data from the Medical $xpenditure %anel &urvey
(M$%&) to analy#e a crosssectional snapshot of the prevalence of diabetes and diabetesrelated cardiovascular
conditions (primary outcome measures) according to activity level and BMI among a nationally representative
population of adults from '((( to '(('. )dditionally, logistic regression analysis was used to estimate the
ad*usted odds of having diabetes and related cardiovascular comorbidities (heart disease, hypertension, and
hyperlipidemia) by activity level (active vs. inactive) and BMI (normal, overweight, obese class I and II, and
obese class III).
Likelihood of diabetes and cardiovascular comorbidities
Health Condition BMI: Normal Obese, class III
Diabetes
Active O1 O2
Inactive
Diabetes & hyperlipidemia OK

"he M$%& household component provided data on selfreported health information (i.e., BMI, physical
activity, covariates) and the M$%& medical provider component provided data on I+,- medical condition
diagnoses.
Main study findings: .verweight, obesity, and inactivity were statistically significantly and independently
associated with increased li/elihood of diabetes and diabetesrelated cardiovascular comorbidities after
ad*usting for sociodemographic characteristics and smo/ing status. In other words, being either overweight,
obese or physically inactive increases the li/elihood of diabetes and related comorbidities.
Lauren Silver HPM 582: Design Analysis #1
Threats to internal and/or external validity: &ince M$%& is based on selfreported data, the estimates of the
prevalence of diabetes and related comorbidities, activity level, and BMI may be under, over, and under
reported, respectively. &ince the data are nationally representative, there are no threats to external validity.
Additional research question: )re there identifiable BMI thresholds at which activity level can help prevent
the onset of diabetes and related comorbidities!
2. Improvements in diabetes processes of care and intermediate outcomes, United States, 19-2002
Objective: "his study sought to examine 0(year changes in the 1uality of diabetes care in the 2.&. using
standardi#ed measures of 1uality, including glycemic control, blood pressure, lowdensity lipoprotein
cholesterol level, and annual foot and dilated eye examinations, etc.
Design and data sources: "his study used crosssectional data from two nationally representative surveys3the
4ational 5ealth and 4utrition $xamination &urvey (45)4$&) and the Behavioral 6is/ 7actor &urveillance
&ystem (B67&&)3to examine secular trends in the 1uality of diabetes care using the 4ational ,iabetes 8uality
Improvement )lliance measures. ,ata on survey participants 09 to :; years old who reported a previous
diabetes diagnosis by health care professional were included in both baseline and followup measurements.
Baseline measurements used 45)4$& data from 0-990--< and B67&& data from 0--; while followup
measurements used 45)4$& data from 0---'((( and B67&& data from '(('. "he primary outcome
measure was the percentage change in individuals who received processes and intermediate outcomes of care
for diabetes from baseline to followup.
Time 1 Time 2
O! O"
Main findings: "he percentage change in individuals who received appropriate care for diabetes from baseline
to remeasurement exhibited improvement in the following process and intermediate outcome indicators= >,>
levels (statistically significant), annual lipid testing, dilated eye and foot examinations, selfmonitoring of blood
glucose level, adoption of aspirin use, and pneumococcal and influen#a vaccinations. Improvement in lipid
control and some improvement in glycemic control was observed. 4o improvement in blood pressure control
was observed, and individuals with )0c ? @A showed a decline in improvement, decreasing by : percent.
Lauren Silver HPM 582: Design Analysis #1
Threats to internal and/or external validity: If under or overreported, the selfreported data from the
45)4$& and B67&& may have resulted in under or overestimates of the percentage change in 1uality
indicators. )dditionally, the large sample si#e of the B67&& may be the cause of statistically significant
findings, rather than actual differences between comparison groups.
Additional research question: Bhat might account for the decrease in the percentage of diabetes patients with
an )0c level less than @ percent occurring simultaneously with a decrease in the percentage of patients with an
)0c level greater than 9 percent!
!. Slo"in# the $ro"th of %ealth &are &osts'(essons from )e#ional *ariation
Objectives: "his commentary draws on prior research to describe possible sources of regional differences in
per capita health care spending as well as strategies for reducing the growth in health care spending.
Design and data sources: "his commentary does not represent a true research design. 6ather, it synthesi#es
data and findings from other research studies (e.g., the ,artmouth )tlas %ro*ect) on variations in health care
spending and spending patterns among 2.&. regions to outline an overall strategy for reducing per capita
spending.
Main findings: 7irst, differences in physician decisionma/ing3as opposed to new technology or payment
systems3most li/ely account for regional differences in per capita health care spending. %hysicians, therefore,
can play a critical leadership role in determining what treatments their patients need and where they should
see/ their care as well as influencing future growth, i.e., choosing conservative vs. highercost treatments,
arguing against health care infrastructure expansions, and supporting costsaving changes in the health care
delivery system. 7inally, payers and policyma/ers should pic/ up where physicians leave off by implementing
policies that encourage highgrowth regions to operate more li/e lowgrowth regions.
Threats to internal and/or external validity: In terms of the suggestions outlined in the commentary, the
author assumes that highgrowth regions would operate li/e lowgrowth regions if properly incentivi#ed via
policy. 5owever, there may be systematic differences between the two types of regions that would prevent this
from occurring, thus, external validity may be a problem.
Lauren Silver HPM 582: Design Analysis #1
Additional research question: )side from physician decisionma/ing, what are other, specific differences
between low and highgrowth regions that account for variations in spending!
+. ,o -uality Improvement Or#ani.ations Improve the -uality of %ospital &are for /edicare
0eneficiaries1
Objectives: "he primary ob*ective of this study was to determine whether the 1uality of care for Medicare
beneficiaries improves more in hospitals voluntarily participating with 8uality Improvement .rgani#ations
(8I.) versus nonparticipating hospitals. ,o participating hospitals demonstrate larger changes in improvement
on 0; 1ualityofcare indicators in five clinical areas= atrial fibrillation, acute myocardial infarction, heart
failure, pneumonia, and stro/e from baseline (0--9) to followup ('((('((0)! )lso, this study sought to
explore characteristics of hospitals that participate with 8I.s versus those that do not.
Design and data sources: "his study was a retrospective pretestposttest, none1uivalent comparison group
1uasiexperimental design3the intervention group consisted of hospitals participating with 8I.s while the
comparison group consisted of nonparticipating hospitals. 5ospitals were considered actively participating with
a 8I. if 0) they used data collected by themselves or by the 8I. for 1uality performance monitoring as a result
of wor/ing with the 8I. andCor ') implemented systems changes as a result of wor/ing with a 8I.. 5ospitals
were considered nonparticipating participating if they did neither of these. "he primary outcome measure was
percent change in patients receiving 1uality indicators from baseline (0--9) to followup ('((('((0).
Main Findings: 2nad*usted analyses demonstrated that participating hospitals improved on 0D of 0; 1uality
indicators while nonparticipating hospitals improved on 0' of 0; indicators. "here were statistically significant
differences on four of 0; 1uality indicators at followup, with participating hospitals performing better than
nonparticipating hospitals on all four. )d*usted analyses demonstrated that while controlling for hospital bed
si#e, profit status, patient age, sex, and race, participating hospitals exhibited a statistically significantly greater
improvement than nonparticipating hospitals on only one of 0; 1uality indicators (patient screened for or given
O1 X O2
O3 O4
#here $! and $% represent baseline meas&rement 'or participatin( and nonparticipatin( hospitals) *
represents the +IO intervention to ,hich participatin( hospitals ,ere e-posed) and O" and O.
represent 'ollo,/&p meas&rement
Lauren Silver HPM 582: Design Analysis #1
pneumococcal vaccine). %articipating hospitals improved more than nonparticipating hospitals on 9 of the
remaining 0< indicators and nonparticipating hospitals improved more than participating hospitals on @
indicators. )dditionally, across all five clinical areas, nonparticipating hospitals were statistically significantly
more li/ely to be smaller and forprofit. )ll in all, participating hospitals do not appear to perform better than
nonparticipating hospitals.
Threats to internal and/or external validity: "his study may be plagued by a number of threats to internal
validity, including history and selection threats. %articipating and nonparticipating hospitals may have been
exposed to external 1uality improvement initiatives that account for improvement in outcomes rather than the
8I. program. %articipating hospitals may differ from nonparticipating hospitals in ways that account for
differences in observed outcomes, such as participating with 8I.s because they are more motivated to improve
1uality or because they have more financial and human resources for 1uality improvement activities.
)dditionally, general secular trends may account for observed outcomes. .nly four 8I.& from five of the ;(
states provided the data for this study, thus, it is 1uestionable whether the results are externally valid.
Additional Research uestion: ,oes the performance of participating hospitals on the 0; 1uality of care
indicators vary depending on the type of technical assistance strategy employed by the 8I.!
2. 3unctional Outcome after 4otal 5nee )eplacement varies "ith 6atient 7ttributes
Objectives: "his study sought to explore patient attributes3gender, age, BMI, emotional health, /nee
diagnosis, 1uadriceps strength, and physical function3associated with varying levels of functional
improvement 0' months following total /nee replacement ("E6) surgery.
Design and data sources: "his study used observational, crosssectional data as part of a preexperimental
onegroup, prettestposttest design. ,ata on preoperative attributes of "E6 patients were obtained from a
national registry that collects demographic and preoperative clinical characteristics, surgeon assessments,
surgical techni1ue, and patientreported outcomes at 0' months after "E6. )ll physicians using Fimmer 4ex
Gen /nee components were eligible to participate in the registry. "he study included data from 9,(;( primary,
unilateral "E6 patients enrolled in the prospective registry between '((( and '((; and who had completed a
full year of outcome data.
Lauren Silver HPM 582: Design Analysis #1
O1 X O2
#here O! represents pre/012 physical '&nction) * represents 012 s&r(ery) and O" represents post/012 physical '&nction 3prettest/
posttest scores e-amined by (ender, a(e, BMI, emotional health stat&s, 4nee dia(nosis, 5&adriceps stren(th, and physical '&nction6
Main findings: .verall, higher age, higher BMI, low baseline emotional health and physical function, and poor
or fair 1uadriceps strength predicted patients who would be classified as belonging to the Glow function
responder group.H %hysical function improved among all age groups, but improved less as the age of the group
increased. "here was no observed difference between gender in functional outcome, however, female function
pre and post"E6 was lower than male function. %atients with a BMI greater than <( were more li/ely to
report lower physical function improvement. %atients with poorer emotional health before "E6 reported less
improvement in physical function post"E6 than patients with a higher preoperative emotional status. More
patients with poor 1uadriceps strength had 0'month %+& scores less than D( than patients with good pre"E6
strength.
Threats to internal and/or external validity= .nly onehalf of total patients in the registry had complete
outcome data, thus, it is possible that these patientsI outcomes differ systematically from those who had
incomplete data. )s a result, this study may not be externally valid. )dditionally, internal validity may be
threatened by history threats. ,ifferences in outcomes between comparison groups may be due to external
events to which patients were exposed rather than differences in attributes, such as the possibility that some
patients participated in rehabilitation programs that were not recorded in the registry.
Additional research question: )mong the patient attributes that affect functional gain following "E6, are
some attributes more important than others in predicting postoperative functional improvement!
8. 9valuatin# &laims-based Indicators of the Intensity of 9nd-of-life &ancer &are
Objectives: "his study sought to= 0) evaluate the accuracy of potential performance measures for endoflife
cancer care using existing administrative data (i.e., Medicare claims and local billing data), ') establish
achievable benchmar/s of care, and D) assess practice variability.
Design and data sources: ,isease and patientrelated data were obtained from 4+IIs &$$6 tumor registries.
,ata from inpatient and outpatient care files, physician and laboratory billings, and bills for home health and
hospice care were obtained from a +M& database. "his study was not conducted according to a GtypicalH
Lauren Silver HPM 582: Design Analysis #1
research design. 6ather, performance measures were assessed in terms of benchmar/s, accuracy, and variability
using various statistical and data mining (e.g., comparing billing claims to medical records) techni1ues.
Main Findings: &ystems in which - out of ten patients do not receive chemo in the last 0< days of life, more
than -9 percent do not start a new chemo cycle in the last D( days prior to death, more than -@ percent do not
have multiple hospitali#ations or emergency room visits or are admitted to the I+2 in the last month of life, and
less than 0: percent die in an acute care institution would be considered health care systems not providing
overly aggressive cancer care. )ll measures demonstrated an accuracy of 9; to -: percent and exhibited
statistically significant variability between health care service areas.
Threats to internal and/or external validity: "here may be threats to external validity given that the
geographic area covered by &$$6 covers only 0< percent of the )merican population and therefore the data are
only GfairlyH representative. )dditionally, the findings may not be applicable to cancer patients who are less
than @; years old.
Additional research question: 5ow does one determine whether the intensive use of aggressive treatment near
the end of life is suboptimal 1uality of care!
9mer#in# ,issertation 4opic 7rea
Main study questions: Both the prevalence and incidence of obesity and related chronic conditions are rising
and have resulted in a public health pandemic of obesity and related comorbidities, including cardiovascular
and pulmonary disease, type II diabetes, cancer, and osteoarthritis, among others. )dditionally, obesity and
related chronic conditions account for a significant share of the rise in health care spending, overall, as well as a
significant amount of employee absenteeism, reduced productivity, and reduced 1uality of life. Given this
context, 0) how can obesity and obesity-related comorbidities be controlled and perhaps even reducedby
focusing on preventive care policy and related interventions/initiatives at the local and state levels! &ince
lifestyle habits most often are formed while in childhood, ') what types of policies are most effective at
reducing obesity among children! It should be emphasi#ed that addressing obesity through policy presents
uni1ue political challenges because it very often is viewed as regulating private behavior3D) so how do you
reach a bipartisan consensus on implementing potentially effective initiatives! )lso, our current health care
Lauren Silver HPM 582: Design Analysis #1
system is much more organi#ed to deal with acute care than with preventive care3<) so how do you modify the
focus of health care delivery accordingly!
!tudy design and "rinci"al data sources: In response to 1uestion number one above, a potential study could
examine the effectiveness of various employersponsored wellness programs aimed at reducing obesity and
related comorbidities among employees. ) 1uasiexperimental, prettestposttest none1uivalent comparison
group design would offer the opportunity to compare outcomes between employees participating in such
programs versus nonparticipating employees. )dditional studies examining the change, if any, in health care
costs incurred by employers offering health insurance coverage to employees could be conducted. %ossible data
sources include data collected directly by employers as part of wellness programs, health plan administrative
data, and data from the 5ealthcare $ffectiveness ,ata and Information &et (5$,I&).
Threats to internal and external validity: )s with all prettestposttest, none1uivalent comparison group
designs, internal validity may be threatened by selection and history threats. $mployees participating in
wellness programs may differ from nonparticipants in ways that would account for observed outcomes more
than the wellness program itself, such as level of motivation. )dditionally, employees may be exposed to
external wellness education from sources other than an employerIs program, thus, program effects may be due
to outside influences rather than to the program itself. $xternal validity may be threatened to the extent that
only local programs are examined, the results of which would not be generali#able beyond the study sites.
#ssues to address in the near$ter%: 0) &pecify the research 1uestion in a manner detailed enough to lend itself
to operationali#ing the study design (i.e., which specific employer wellness program(s) will be examined,
which employees comprise the comparison and intervention groups, what are the primary outcome measures,
etc.) and ') determine from where and how data will be collected.
:hat is ;or is not< :ron# "ith this 6icture ;,rembo"s=i, 3i#ure 2.2<1
"he location of the labels J)I and J+I should be swapped to convey the correct progression of receiving fewer
to more health services, which leads an individual from suboptimal to optimal health () to B), and then
decreasing increases in level of health with additional services (B to +).