Health informationexchangeacrossus healthinstitution (1)

Running head: 1
Factors affecting Health Information
Exchange across US Health Institutions
University of Illinois
Under the guidance of:
Dr. Ranganathan Chandraskaren
May 3, 2017
Submitted By:
Aadish Chopra
Sahil Sahni
Vivian Lin

2
Contents
1. Overview -------------------------------------------------------------------------------3
2. Dataset----------------------------------------------------------------------------------4
3. Model for Hospital andAmbulatory(in/out)-----------------------------------7
4. Results-----------------------------------------------------------------------------------8
5. Model for HIE patient---------------------------------------------------------------12
6. Results----------------------------------------------------------------------------------13
7. Appendix-------------------------------------------------------------------------------19
8. References-----------------------------------------------------------------------------20

3
Overview
Background:
The data was obtained from the American Health Association. The HHS Office of the National
Coordinator (ONC) partnered with the AHA to track hospital adoption and use of EHRs and the
exchange of clinical data.
The data was obtained via sending a survey (November 2013 – February 2014) to US hospitals,
including non-AHA-member hospitals. Surveys were addressed to the CEO and responses were
submitted by the person most knowledgeable about the hospital’s health IT.
The IT Supplement database provides hospital-level details on EHR adoption, the functionality
of hospitals on meaningful use, health information exchange, and patient engagement, along
with details on EHR system and vendor characteristics.
The response rate was 59% among non-federal acute care hospitals.
The survey results provide the most current and reliable data on IT implementation available.
Objective:
With the use of EHR being mandated across health institutions in the US, the next step is to
improve the level of health information exchange across the institutions. The study seeks to
analyze the factors that affect the Health Information Exchange amongst various hospitals and
health institutions in US.
Methods:
The dataset obtained from the American Health Association provides information regarding the
level of EHR and HIE implementation across various health institutions across United States.
This data was processed and analyzed using linear and logistic regression model to observe the
factors affecting the exchange of health information across the institutions.
A separate analysis was performed to identify factors affecting HIE within the network, outside
the network and the level of health information access provided to the patients

4
Health Information Exchange across US Health Institution
Dataset: Exploratory Data Analysis
Characteristics of Dataset
The given dataset originally had 3284 rows and 171 dimensions. The dataset didn’t had any
errors. Some values were missing but it only suggests that answers were not responded properly,
not that data was collected poorly. Overall 90% of the data was in a good condition.
Data Transformation
The dataset under study had to be transformed in correspondence to the AHA_Schema. Most of
the variables were transformed into binary values since the data that was captured had this
property. Variables that were not transformed were for eg. AHA ID, Zip code, Bed Size etc. For
a complete list check, AHA_Cleaned_Data. For better analysis some variables like Bed Size was
converted to a categorical variable “Large”,” Medium” and “Small” where Large is greater than
150 and Small is lesser than 50.
Data summing
Transformed dataset was then scrutinized and some variables were dropped like Zip Code,
Mechanism involved etc. Variables which were used in the analysis can be found in the
Analysis_Sheet. Selection of variables and assigning values was done based on research. For
example, Patient Demographics is captured as “Fully implemented across all units”,” Fully
implemented in at least one unit”,” Have resources to implement in the next year”,” Do not have
resources but considering implementing” and “Not in place and not considering implementing”.
To reduce this variable analysis was done using different survey techniques such as simple
summing, weighted summing, unique sum and then one versus none. The best technique for this
example was found to be one versus none keeping in mind the question that is to be answered.
Data Pre-processing Step
Checked the missing values across the dependent variables that are with hospital outside your
system (HOS) and with ambulatory providers outside your system (AOS). From the table above,
283 records were deleted because they were missing across all the variables. It was noticed that
there were 4 missing ID in our dataset, thus, the entire records were deleted.
Scores were assigned to all variables (check schema titled AHA_Schema). Summing of scores
was done to calculate with hospital outside your system (Hospital outside system) and with
ambulatory providers outside your system (Ambulatory outside system) and independent
variable EHR score.
Statistical Technique
Based on the given dataset and the probing question “What factors affect Healthcare information
exchange within and across health institutions?” the best technique was found to be Linear
Regression. Principal Component Analysis was done but only to reduce the number of

5
dimensions on selected variables. Anova was done to determine if there is a significant effect of
Bed Size.
Fig 1: Count of missing values in the dataset
In MEANS Procedure table, it can be inferred that Do not exchange patient data variables are
problematic because they contain too many missing values. Therefore, we dropped these
variables from our further analysis.
Missing Value Analysis
Missing values are same for Hospital out, Hospital in and Ambulatory in and Ambulatory outs
respectively i.e (m=283) values.

6
Fig 2: Total number of rows having missing values in dependent variables
Top three vendors analysis is in the appendix excel sheet. The top three vendors are Cerner,
Epic, and Meditech. Based on results found new variables were created topVendor_in,
topVendor_out.
Descriptive Statistics
Variable N Mean Std Dev Sum Minimum Maximum
EHR score 3284 42.63 16.13 139963 0 62
Depout_Hospital 2995 1.92354 2.47401 5761 0 6
Depout_Ambulatory 2995 2.55693 2.50857 7658 0 6
Depin_Hospital 3284 3.612549497 2.742290893 11862 0 6
Depin_Ambulatory 3284 3.613158696 2.607862739 11860 0 6
BedSmall 2995 0.29616 0.45664 887 0 1
BedMed 2995 0.24441 0.42981 732 0 1
BedLarge 2995 0.45943 0.49843 1376 0 1
Upfront capital costs 2820 0.57447 0.49451 1620 0 1
Ongoing costs of
maintaining/upgrading 2817 0.62939 0.48305 1773 0 1
Obtaining physician cooperation 2817 0.57685 0.49415 1625 0 1
Obtaining other staff cooperation 2817 0.26376 0.44075 743 0 1
Concerns about security/liability for
privacy breaches 2817 0.23642 0.42496 666 0 1
Uncertainty about certification
process 2817 0.16897 0.3748 476 0 1
Lack of vendor capacity 2816 0.23651 0.42501 666 0 1
Lack adequate hospital IT staff 2816 0.47372 0.4994 1334 0 1
Meeting all meaningful use criteria
on time 2815 0.62131 0.48515 1749 0 1
Difficulty reaching system-level
decision 2815 0.24334 0.42917 685 0 1
topVendor_in 2995 0.50284 0.50008 1506 0 1
topVendor_out 2995 0.38965 0.48775 1167 0 1
Redesignedworkflows 2778 2.38013 0.85679 6612 0 4
Table 1: Descriptive statistics for the independent and dependent variables

7
Top three vendors were also calculated based on market research for the year 2013 and the same
result was found out. There is significant difference in the Hospital out and Hospital In as well as
Ambulatory In and ambulatory Out. The three main important barriers in exchanging Health
information are “meeting meaningful use”, “obtaining costs of maintaining/upgrading” and
“obtaining physician cooperation”. Average EHR score is 42.63 which is high even when there
are 1023 small sized and 822 medium sized hospitals
Model for Hospital and Ambulatory(in/out)
In this 2013 America Healthcare Associate Annual Survey Questionnaire, we aim to evaluate
healthcare information exchange inside and outside system across hospitals. Therefore, two
models were built with different provider types as dependent variables which are with hospital
outside your system (for Hospital outside system) and with ambulatory providers outside your
system (for Ambulatory outside system). For the independent variables, we grouped the
questions into five main factors which are Hospital Characteristics, EHR usage, Vendor types,
Workflows Redesigned, and Barriers.
We consider bed size as Hospital Characteristics, EHR score as EHR usage, top three vendors as
Vendor type, workflow redesign level as Workflow redesigned and Upfront capital costs,
Fig3: Model representing independent variables (clubbed together for better visualization) and dependent variables.
Dependent variables shown only for Hospital outside and Ambulatory outside.
Ongoing costs of maintaining/upgrading, obtaining physician cooperation, Obtaining other staff
cooperation, Concerns about security/liability for privacy breaches, Uncertainty about
certification process, Lack of vendor capacity, Lack adequate hospital IT staff, Meeting all
meaningful use criteria on time and Difficulty reaching system-level decision as Barriers.

8
Results
Regression Analysis for all hospitals
. p-value < 0.1
* p-value < 0.05
** p-value < 0.01
*** p-value<0.001
Hospital Inside and Ambulatory Inside
Critical comments
Although the survey was filled by a responsible person there are contradictions in some parts of
the survey. For example, survey asks a question about what type of EMR system is used. People
who have marked “Do not know” have marked “Fully implemented across all units” in
dependent variable “Patient Demographics”,” Laboratory Results” etc. Variables which are
conflicting each other have been dropped
Hospitals which have low number of beds are scoring extremely well in EHR score. The result is
contrary to what people think that more the number of beds more likely it is to adopt technology
Filtering the dataset based on bed category. There are 1023 small sized hospitals (categories have
been set based on market research and internet) ,822 Medium and rest are large sized categories.
Table Numberof Hospitals Mean EHR Score Standard DeviationEHR score
Small 1023 37.66 16.13
Medium 822 39.51 16.13
Large 1439 47.91 16.13
Table 2: Breakdown of hospitals according to bed size v/s Mean EHR score

9
Fig 4: Scatterplot between Number of years having EMR against EHR score
The above scatterplot shows that even when EMR has been adopted late EHR score is high.
From this scatterplot two meaningful results can be inferred 1) EMR’s were put in place to
implement healthcare information exchange. 2)EMR’s which were already in place were either
underutilized or there was no system for health information exchange.
Results from Regression
 Institutions tend to do less of HIE if that is a small or a medium
 More the EHR score, more is the HIE.
 Workflows redesigned is an important parameter in deciding the level of HIE.
Conflicting results
 Obtaining staff cooperation, complexity associated with coordinating and uncertainty are
going in the opposite direction if compared to Hospital In. It means more staff
cooperation less is HIE which is conflicting with common sense.
 Challenge/complexity of meeting requirements on time and Obtaining staff cooperation
are going in opposite direction to Ambulatory out. It means more complexity more is HIE
which is again contrary to common sense

10
Hospital Outside and Ambulatory Outside
The goal is to know what factors affect healthcare information exchange outside and inside the
system.
For Hospital outside system model, the important factors identified were and written in the
importance order as below: EHR score, top Vendor, obtaining other staff cooperation, lack
adequate hospital IT staff, bed size, Obtaining physician cooperation and Lack of vendor
capacity. Here, I would like to explain one of the results from my variables. We can see the p-
value for Obtaining other staff cooperation is 0.01 statistical significant and has -0.06
standardized coefficients. This means the more challenge the hospital obtains other staff’s
cooperation, the less they meet the meaningful use in implementing EHR. Thus, lessen in
exchanging healthcare information outside the system.
For Ambulatory outside system model, the important factors identified were EHR score, top
Vendor, bed size, lack adequate hospital IT staff, and Meeting all meaningful use criteria on
time. Here, I would like to explain one of the results from my important variables. We can see
the p-value for Lack adequate hospital IT staff is 0.01 statistical significant and has -0.05
standardized coefficients. This means the more challenge the hospital get IT support, the less
they meet the meaningful use in implementing EHR. Thus, lessen in exchanging healthcare
information outside the system.
Furthermore, since we know bed size affect health information exchange outside the system
based on statistical significant results above, we then want to understand whether important
variables vary across different bed sizes
Note: The dependent variable remains the same and all the bed size models include Hospital
Characteristics, EHR usage, Vendor type, Workflow redesign, and Barriers variables.

11
Models
Hospitals in Ambulatory in Hospitals outAmbulatory out
Adj R-squre 0.1773 0.2215 0.082 13.39
F-Value 41.67 54.74 12.65 21.87
Hospital Characteristics
Bed Size
Small ***-2.97 ***-3.22 NS -0.10562
Medium ***-0.13 ***-0.14 **-0.06504 -0.08916
Large(ref)
Level of EHR Usage
EHR score ***0.12 ***0.014 ***0.18722 0.267
Vendor Type
Is it a Top Inpatient EHR Vendor NS NA ***0.10457 NA
Is it a Top Outpatient EHR Vendor NA NS NA 0.0768
Barriers
Upfront capital costs/lack of access to capital to install systemsNS NS NS NS
Ongoing cost of maintaining and upgrading systemsNS NS NS NS
Obtaining physician cooperation ***-0.07 NS *-0.05952 NS
Obtaining other staff cooperation *0.04 **0.05 ***0.08137 NS
Concerns about security or liability for privacy breachesNS *-0.03 NS NS
Uncertainty about certification requirements*0.04 NS NS NS
Limited vendor capacity NS **-0.05 .-0.03972 NS
Lack of adequate IT personnel in hospital to support implementation/maintenanceNS NS ***-0.07238 -0.05264
Challenge/complexity of meeting all meaningful use criteria within implementation timeframeNS ***0.05 NS 0.06011
Complexity associated with coordinating decision with system-level leadership***0.13 NA NS NS
Workflow Redesigned ***0.03 **0.018 NS NS
Table 3: Regression results for Hospital In,Ambulatory In,Hospital Out anf Ambulatory Out
HIE Patient
Model:
There were eight different functionalities that were asked via the survey questionnaire to get
insights regarding the level of health information exchange a health institution practices with
their patients.
All these variables were captured as binomial variables (1 if the health institution offers the
feature to the patients and 0 if the feature is not provided). Hence, we performed a logistic
regression analysis on all these variables to identify the factors affecting these variables.
For the independent variables, we grouped the questions into five main factors which were
expected to affect health information exchange.
 Hospital Characteristics

12
 EHR usage
 Vendor types
 Workflows Redesigned
 Barriers.
Fig5: Model representing independent variables (clubbed together for better visualization) and dependent variables.
Dependent variables shown only for HIE patient.
Results:
Independent Variables
View informationfromtheirhealth/medical records
online
(Intercept) -0.364598*
`Bedcategory`Medium 0.030977
`Bedcategory`Small 0.114208
`Level ofEHR` 0.001679
`Hospital IN` -0.003938

13
`Hospital Out` 0.007868
ÀmbulatoryIN` 0.030143
ÀmbulatoryOut` -0.027843
ÈEHR Vendor` 0.118615*
Ùpfrontcapital costs, or lack of capital` 0.007244
Òngoingcosts of maintaining/upgrading` -0.149825
Òbtainingphysiciancooperation` -0.121053
Òbtainingother staffcooperation` 0.163066
`Concernsabout security/liabilityforprivacy breaches` 0.074162
Ùncertaintyabout certificationprocess` -0.134643
`Lack of vendor capacity` 0.113308
`Lack adequate hospital IT staff` 0.007129
`Meetingall meaningful use criteriaon time` 0.034661
`Difficultyreachingsystem-level decision` -0.025348
`Redesignedworkflowstomake optimal use of EHR` -0.016866
`Numberof Yearshaving EMR` 0.016810*
Request an amendment to change/update
their health/medical record
(Intercept) -0.620901 ***
`Bedcategory`Medium -0.063008
`Hospital IN` -0.030569 .
ÈEHR Vendor` 0.08294
Òngoingcosts of maintaining/upgrading` -0.217381 *
`Concernsabout security/liabilityforprivacy breaches` -0.076554

14
`Difficultyreachingsystem-level decision` 0.000942
`Redesignedworkflowstomake optimal use of EHR` 0.004442
`Numberof Yearshaving EMR` 0.004832
Download information from their health /
medical record
(Intercept) -0.951185***
ÀmbulatoryOut` -0.043078.
ÈEHR Vendor` 0.093394.
Ùpfrontcapital costs, or lack of capital` -0.033897
Òngoingcosts of maintaining/upgrading` -0.178963.
Òbtainingother staffcooperation` 0.228925*
`Lack adequate hospital IT staff` -0.02342
`Numberof Yearshaving EMR` 0.014829*
Patientscanelectronicallytransfertoa thirdparty
(Intercept) -2.347344***
`Bedcategory`Small -0.05638
ÀmbulatoryIN` -0.027657
ÀmbulatoryOut` 0.012433
ÈEHR Vendor` -0.018728

15
`Numberof Yearshaving EMR` 0.017500.
Requestrefillsforprescription
(Intercept) -0.748336 ***
`Level ofEHR` -0.00064
`Ongoingcosts of maintaining/upgrading` -0.198478 .
`Obtainingother staffcooperation` 0.209446 .
`Numberof Yearshaving EMR` 0.013249 .
Schedule appointmentsonline

16
(Intercept) -1.087584***
`Hospital Out` -0.014552
ÈEHR Vendor` 0.156488**
`Numberof Yearshaving EMR` 0.01132
Pay bills online
(Intercept) -0.320766 .
`Hospital IN` -0.032905 .
ÀmbulatoryIN` -0.015983
ÈEHR Vendor` 0.147514 **
Òbtainingother staffcooperation` -0.003594

17
`Numberof Yearshaving EMR` 0.040680 ***
Submitpatient-generateddata
(Intercept) -1.615576 ***
`Bedcategory`Small -0.156652
`Level ofEHR` -0.002404
`Hospital Out` -0.003018
ÀmbulatoryOut` 0.000135
Òngoingcosts of maintaining/upgrading` -0.318555 *
Òbtainingphysiciancooperation` -0.379243 **
`Concernsabout security/liabilityforprivacy breaches` 0.278141 *
Ùncertaintyabout certificationprocess` -0.364821 *
`Numberof Yearshaving EMR` 0.030971 ***
Table 4: Logistic regression Results for HIE Patient
From the above analysis, we observe that:
 If the health institution opted for one of the top 3 EHR vendors in the market place for
inpatient and outpatient, the more likely they are to allow the patient to view information
from their health/medical records online, Schedule appointments online and Pay bills
online
 The more the number of Years the hospital has had EMR active within their institution,
they are more likely to engage in health information exchange with their patients
 We also observe that their health information exchange is not affected significantly with
their level of EHR implementation.

18
 The level of HIE within the network and outside the network affects the HIE patient for
large hospitals only. The more the level of HIE within the network for large hospitals, the
less they are likely to allow patients to Submit patient-generated data
 The more the ongoing costs of maintaining/upgrading, the less likely the hospital is to
allow patients to Submit patient-generated data or Request an amendment to
change/update their health/medical record
 Obtaining physician cooperation proves to be very crucial to provide patients with the
opportunity of providing patient generated data online. The higher difficulty in obtaining
physician cooperation for EHR implementation, the less likely they are to provide
patients a portal to submit the patient generated data

19
Appendix
Following files are attached
1) 2013 AHA Annual Survey IT Supplement Documentation.pdf
2) AHA Schema HIM.xlsx
3) AHA_Cleaned Data.xlsx
4) Correlation_Matrix.xlsx
5) HIE AHA Data Output.xlsx

20
References
1) Walker, J., Pan, E., Johnston, D., & Adler-Milstein, J. (2005). The value of health care
information exchange and interoperability. Health affairs, 24, W5.
2) Miller, A. R., & Tucker, C. (2014). Health information exchange, system size and
information silos. Journal of health economics, 33, 28-42.
3) Mobile Communication Between Patients and Providers: An Examination of Patients’
Willingness to Exchange Health Information Using Mobile Devices by Dr. C.
Ranganathan
4) www.aha.org/research/rc/stat-studies/fast-facts.shtml
5) searchhealthit.techtarget.com/feature/Who-are-the-top-EHR-vendors-in-your-state
6) https://www.healthfusion.com/ehr-features/ehr-vendor-comparison

Health informationexchangeacrossus healthinstitution (1)

More Related Content

What's hot (19)

Similar to Health informationexchangeacrossus healthinstitution (1) (20)

More from University of Illinois,Chicago (15)

Recently uploaded (20)

Health informationexchangeacrossus healthinstitution (1)