Keywords
blood glucose self-monitoring - self-management - patient portals - diabetes - obstetrics–gynecology
Background and Significance
Background and Significance
Patient-generated health data (PGHD), that is, data recorded or generated by patients
during their daily lives, could give clinicians a much more comprehensive understanding
of patient health status than is available from clinical visits alone. It seems likely
that a variety of types of PGHD could enrich clinical care, including exercise, activity,
or diet tracking, through wearables and mobile devices[1] or through self-tracked data from personal medical devices such as blood pressure
and blood glucose monitors.[2]
[3] Engaging patient in tracking and sharing PGHD could be a way to promote health knowledge,
disease self-management skills, and motivation, all of which are known to improve
health outcomes.[4]
[5]
[6]
Some important barriers stand in the way of integrating PGHD into clinical practice.
The use of computers, smartphones, and other health information technology remains
lowest among the patients in the greatest need, including people who are elderly,
less educated, or less affluent.[7]
[8]
[9]
[10] Although some communities embrace PGHD and data tracking, other people with chronic
medical conditions may be deterred from self-tracking by feelings of frustration or
anxiety in response to their PGHD data, which can serve as an unwelcome reminder of
their poor health.[11] On the provider side, physicians who have expressed interest in potentially using
PGHD (for example, in diabetes)[12] still require additional support before these data are likely to be useful. Providers
are likely to need practice protocols to guide clinical responses to the data, visualizations,
data reduction, or decision support to help make sense of the information while preventing
overload, electronic health record (EHR) integration, and integration into clinical
workflow.[3]
[12]
[13] Nationality studies, feasibility studies, and demonstration projects to integrate
PGHD with EHRs are ongoing.[14]
[15]
Prior to these national studies, in 2012, our academic multispecialty practice sought
to facilitate the use of PGHD by enabling an electronic data tracking tool called
a flowsheet. The functionality was made available in the EHR where it could be adopted
by providers as needed, but there was no specific implementation plan, encouragement,
or incentivization to adopt this function. In this retrospective study, we sought
to describe adoption rates and characteristics of early adopters of the PGHD functionality
under these naturalistic conditions, along with preliminary data about associations
with patient outcomes.
Objectives
In this retrospective study, our objectives were to describe adoption rates and characteristics
of early adopters of the PGHD functionality under these naturalistic conditions and
to collect preliminary data on associations with patient outcomes.
Methods
Setting and Technology
The Weill Cornell Physician Organization is the multispecialty faculty practice of
Weill Cornell Medical College in New York City, representing both physician ambulatory
offices and ambulatory hospital-based clinics. All physicians use the EpicCare electronic
medical record and its integrated portal, which is branded as Weill Cornell Connect.
Weill Cornell physicians see approximately 300,000 unique patients per year. Of the
patients, 42% have electronic patient portal accounts. Providers are not salaried
by Weill Cornell and participate in both public and private fee-for-service and managed-care
plans. (Weill Cornell is affiliated to NewYork-Presbyterian Hospital, the largest
hospital system in New York City, but inpatient electronic systems and databases are
separate from the outpatient ones and are not described in this paper.)
The flowsheet functionality was enabled in 2012 and was available to every provider.
A provider selects patients for whom it appears appropriate and introduces the flowsheet
to them; patients cannot initiate the use of flowsheets by themselves. After a physician
enables the flowsheet for a patient, the patient can use it to upload data securely
through the electronic patient portal that is accessible either through a web browser
or a smartphone app. Each blood glucose value is labeled with the time, and the patient
can also enter insulin dose (if any), time of the insulin administration, and free
text notes. Patients may upload up to several values per day or less frequently. The
physician can then view the data within the EHR in a spreadsheet format.
When this flowsheet functionality was enabled, announcements went out to all EHR users,
but the tool was not attached to any specific implementation plans, research studies,
or quality initiatives.
Methods
We defined flowsheet adopters as adult patients who (1) had uploaded three or more
blood glucose values within any 9-month period, (2) had at least a 2-year history
at Weill Cornell before the first data upload, and (3) had logged into their patient
portal account at least once before the first data upload. The comparison group comprised
patients who had encounters with any of the physicians seen by the flowsheet adopter
patients during the same time period, had any recorded diagnosis of diabetes or gestational
diabetes, and had a 2-year history as well as at least one portal log-in before the
date defined as the index date (see definition below). The 2-year history at Weill Cornell was chosen to ensure
that there was comparable run-in data on diagnoses, clinical visits, and portal use
for both flowsheet adopter and comparison patients. All variables were specified a
priori, with the exception of the three-value upload threshold that was developed
after inspecting the data.
The index date for the flowsheet adopter group was the date of first PGHD data upload. For the comparison
group, the index date was set as the median time from between the first portal log-in and December 2016.
We assessed physiological indicators (hemoglobin A1c, blood pressure, low-density
lipoprotein cholesterol, and body mass index [BMI]) before and 9 months after the
index date. Nine months was chosen to provide sufficient time for two to three hemoglobin
A1c measurements.
The Johns Hopkins Aggregated Diagnostic Group algorithm, a case mix metric appropriate
for ambulatory patients, was run on patient diagnosis codes to produce a chronic condition
count.[16] Patient addresses were mapped to census block to estimate patient socioeconomic
status using the Centers for Disease Control's Social Vulnerability Index–Socioeconomic
Status (SVI-SES) theme, a composite metric of neighborhood-level income and education
levels.[17] Values range from 0 to 1, with 1 indicating the highest degree of vulnerability.
Statistical Analysis
Categorical data were compared using chi-square tests (or Fisher's exact when cell
sizes were smaller than 5). Continuous and count data were compared using independent
sample t-tests when normally distributed or Wilcoxon's tests when skewed. Analyses were conducted
using SAS v. 9.4 (SAS Institute Inc., Cary, North Carolina, United States) and R v.
3.5.0 (R Foundation, Vienna, Austria).
Results
All providers had access to the flowsheets over 4 years, including approximately 250
attending physicians in internal medicine (including endocrinologists) and approximately
40 attending physicians in obstetrics and gynecology (ob-gyn). During these 4 years,
37,578 patients had a recorded diagnosis of diabetes, 8,851 (23.5%) of whom had portal
accounts. Also, during this time, 1,395 patients had gestational diabetes, 823 (59%)
of whom had portal accounts.
Over 4 years, 16 providers chose to use the blood glucose flowsheet, and 12 of them
had patients who met our inclusion criteria. Of these 12, 4 were ob-gyn physicians
(representing approximately 10% of the ob-gyn with access to the flowsheets), and
4 were physicians in the Department of Medicine (representing ∼2% of the internists
with access to flowsheet). One of the internists had a specialty of endocrinology.
The remaining providers who used the flowsheets were nurses in the Departments of
Obstetrics and Gynecology (n = 2) and Medicine (n = 1), one individual with unknown credential in the Department of Obstetrics and
Gynecology, and one individual missing both department and credential.
Fifty-three (53) patients uploaded three or more blood glucose values over any 9-month
period. Of these patients, 23 were pregnant women (representing ∼3% of portal users
with gestational diabetes) and 30 were nonpregnant adults with diabetes (representing
<1% of portal users with diabetes). A total of 201 nonpregnant comparison patients
and 41 pregnant comparison patients were identified.
Pregnant Patients
The 23 pregnant patients who uploaded PGHD submitted a median of 201 (range: 23–573)
values to six providers in the Department of Obstetrics and Gynecology. Uploaders
were similar to comparison group patients in race, ethnicity, age, and socioeconomic
status assessed through both SVI-SES and insurance type ([Table 1]). However, uploaders had more clinical encounters and portal log-ins before initial
data upload, somewhat earlier establishment of patient portal accounts, and worse
baseline blood pressure.
Table 1
Characteristics of patient-generated health data uploaders and comparison with patients
who did not upload
|
|
Chronic disease patients
|
Pregnant women
|
|
|
Uploaders (n = 30)
|
Comparison patients (n = 201)
|
p-Value
|
Uploaders (n = 23)
|
Comparison patients (n = 41)
|
p-Value
|
|
Race
|
White
|
14
|
46.7%
|
119
|
59.2%
|
0.05
|
14
|
60.9%
|
24
|
58.5%
|
0.94
|
|
Black
|
1
|
3.3%
|
8
|
4%
|
|
0
|
0%
|
0
|
0%
|
|
|
Asian
|
7
|
23.3%
|
24
|
11.9%
|
|
6
|
26.1%
|
10
|
24.4%
|
|
|
All other
|
8
|
26.7%
|
50
|
24.9%
|
|
3
|
13%
|
7
|
17.1%
|
|
|
Ethnicity
|
Hispanic
|
4
|
13.3%
|
16
|
8%
|
0.28
|
2
|
8.7%
|
4
|
9.8%
|
>0.99
|
|
Not Hispanic
|
24
|
80%
|
146
|
72.6%
|
|
20
|
87%
|
34
|
82.9%
|
|
|
All other
|
2
|
6.7%
|
39
|
19.4%
|
|
1
|
4.3%
|
3
|
7.3%
|
|
|
Age
|
|
39.2 (14.4)
|
47 (15.2)
|
0.004
|
31.1 (4.5)
|
29.9 (4.5)
|
0.37
|
|
Payer
|
Commercial
|
26
|
86.7%
|
136
|
67.7%
|
0.48
|
20
|
87%
|
36
|
87.8%
|
0.79
|
|
Medicaid
|
0
|
0%
|
3
|
1.5%
|
|
0
|
0%
|
1
|
2.4%
|
|
|
Unknown
|
1
|
3.3%
|
16
|
8%
|
|
3
|
13%
|
3
|
7.3%
|
|
|
Medicare
|
3
|
10%
|
45
|
22.4%
|
|
0
|
0%
|
0
|
0%
|
|
|
Chronic condition count
|
5.9 (4.7)
|
6 (4.5)
|
0.92
|
2.4 (1.9)
|
1.1 (0.9)
|
0.10
|
|
SVI-SES
|
|
0.27 (0.30)
|
0.24 (0.26)
|
0.63
|
3 (2.6)
|
1.9 (2.2)
|
0.62
|
|
Encounters year before index
|
10.2 (4.6)
|
6.5 (5.3)
|
<0.001
|
14 (9.1)
|
8.1 (6.1)
|
0.003
|
|
Portal log-ins year before index
|
65.9 (47.6)
|
23.5 (28.1)
|
<0.001
|
65.4 (44.7)
|
26.2 (34.3)
|
<0.001
|
|
Year of first encounter
|
2,008 (3.3)
|
2,008 (3.1)
|
0.18
|
2,009 (3)
|
2,010 (2.7)
|
0.08
|
|
Year established portal account
|
2,012 (1.6)
|
2,012 (1.9)
|
0.10
|
2,013 (1.9)
|
2,014 (1.6)
|
0.004
|
|
Baseline hemoglobin A1c (%)
|
7.8 (3.1)
|
6.8 (1)
|
0.17
|
5.5 (0.3)
|
5.7 (5.2)
|
0.54
|
|
Baseline systolic blood pressure
|
117.2 (13)
|
118.5 (16)
|
0.64
|
109 (10)
|
104.1 (6.6)
|
0.04
|
|
Baseline diastolic blood pressure
|
72.3 (10.5)
|
71.8 (9.2)
|
0.82
|
66.7 (6.8)
|
65 (6.7)
|
0.33
|
|
Baseline LDL
|
118.5 (38.1)
|
101.1 (33.2)
|
0.06
|
102.8 (36.3)
|
102 (27.3)
|
0.95
|
|
Baseline body mass index
|
29.2 (6.8)
|
28.2 (6.2)
|
0.46
|
27.2 (5.6)
|
24.7 (3.8)
|
0.06
|
|
Change in hemoglobin A1c (%)
|
–0.8 (1.9)
|
0.1 (0.8)
|
0.02
|
NA[a]
|
NA[a]
|
–
|
|
Change in systolic blood pressure
|
5.5 (16.7)
|
1.6 (16.1)
|
0.46
|
7.3 (8.3)
|
0.4 (9.7)
|
0.04
|
|
Change in diastolic blood pressure
|
3.8 (10)
|
0.9 (9.4)
|
0.38
|
5.8 (10.1)
|
–2.6 (7.4)
|
0.02
|
|
Change in LDL
|
2.1 (33.5)
|
0.1 (23.9)
|
0.28
|
NA[a]
|
NA[a]
|
–
|
|
Change in body mass index
|
–1.4 (2.2)
|
0.2 (1.6)
|
0.01
|
–3.1 (3.2)
|
0.8 (2.6)
|
0.003
|
Abbreviations: LDL, low-density lipoprotein; NA, not available; SVI-SES, Centers for
Disease Control's Social Vulnerability Index–Socioeconomic Status.
a As these indicators are not typically monitored multiple times during pregnancy,
fewer than 10 pregnant patients had both baseline and follow-up values for LDL and
hemoglobin A1c.
Note: p-values of 0.05 or lower are shown in boldface.
For pregnant patients, average BMI dropped significantly more in the 9 months after
the index date (i.e., post-partum) among uploaders than nonuploaders. However, average
blood pressure increased more among uploaders than nonuploaders.
Chronic Disease Patients
A total of 30 nonpregnant patients with diabetes diagnoses uploaded a median of 139.5
(15–1,253) values to 10 providers in the departments of Internal Medicine, Endocrinology,
Cardiology, and Nutrition.
Among these chronic disease patients, uploaders were more likely than comparison group
patients to be Asian-American and were younger, but the groups did not have other
significant demographic differences including in socioeconomic status ([Table 1]). Uploaders had more clinical visits and portal log-ins before initial data upload.
During the 9-month period after the first glucose value upload, uploaders had significantly
larger reductions in hemoglobin A1c and BMI than did nonuploaders ([Table 1]).
Uploaders also experienced large variations in hemoglobin A1c level after the index
date ([Fig. 1]). Uploaders appeared to fall in two subsets ([Fig. 2]). One subset had low, well-controlled hemoglobin A1c values both before and during
PGHD upload. A second subset began upload at a time when their hemoglobin A1c was
elevated and typically experienced a hemoglobin A1c decrease followed by a plateau.
Fig. 1 Average change in hemoglobin A1c values among chronic disease patients who did and
did not upload glucose data. Chronic disease patients who uploaded patient-generated
health data had more variable hemoglobin A1c values over 9 months than comparison
patients. The index date in the graph is the date of the first upload for uploaders
and the median time since initial portal log-in for comparison patients.
Fig. 2 Hemoglobin A1c (HA1c) values among chronic disease patients who uploaded patient-generated
health data. Among chronic disease patients, uploaders were clustered into two groups.
The first (bottom) had well-controlled HA1c values at the time of the first upload
(index date) and maintained low values throughout. The second began uploading at a
time of HA1c elevation and, with one exception, saw reductions after beginning to
upload.
Discussion
Self-tracked or PGHD appears to offer the possibility of improved clinical monitoring
for patients between clinical visits. Many patients who track data would like their
physicians to review it to identify health problems or reassure them that all is well.
Nevertheless, in an academic multispecialty practice with high rates of patient portal
use, adoption of a PGHD data upload function has been slow. Over 4 years, 16 providers
tried it, and 53 established patients uploaded three or more values.
The most marked difference between uploaders and comparison patients was that uploaders
had more visits and portal log-ins. However, uploaders did not have more chronic conditions
than nonuploaders and did not have markedly worse baseline physiological indicators
(with the exception of higher systolic blood pressure among the pregnant subgroup).
Overall, this suggests that uploaders were not substantively sicker than nonuploaders
but rather more engaged in their health care, motivated to change behavior, or loyal
to the medical center. They may also have been more facile with information technology,
a point supported by the fact that uploaders among the chronic disease patients were
younger than nonuploaders. Patient portal use has been shown to be associated with
socioeconomic status, with more affluent patients having higher average health literacy
and better access to computers, broadband internet, and smartphones.[8]
[18]
[19]
[20]
[21] However, in our study, we included only patients with portal accounts and a history
of using them. Among these patients, there were no socioeconomic differences between
uploaders and nonuploaders, as indicated by insurance type or a composite measure
of U.S. Census tract-level income and education.
Among the chronic disease patients, PGHD uploading was associated with marked variation
in hemoglobin A1c values ([Fig. 1]) and significant reductions in hemoglobin A1c and BMI over a 9-month period ([Table 1]). It appears that uploading was associated with a period of medical transition or
change. However, given the baseline differences between the two groups, it would be
difficult to conclude that the uploading behavior caused the reduction.
Among pregnant patients, BMI is expected to rise during pregnancy, but we found that
after birth, BMI dropped more among uploaders than nonuploaders. By contrast, blood
pressure increased more among uploaders than comparison patients. This could be an
artifact of the fact that PGHD patients were slightly sicker (more than two comorbidities
compared with approximately 1; p = 0.1) or that some patients with gestational diabetes may have had comorbid preeclampsia.
Hemoglobin A1c is not typically monitored during pregnancy, and therefore there were
little data on this end point. Again, the baseline differences between the two groups,
especially in terms of patient portal use, mean that the observed effects were associated
with the flowsheet but may not have been caused by it.
The low adoption rate for PGHD upload is probably attributable to several factors.
The first factor is barriers to provider uptake. It is widely accepted that integration
into the EHR is a prerequisite for the use of novel tools, and, in other situations,
lack of integration is cited as a barrier to adoption.[3]
[22] However, even though this PGHD upload tool was integrated into the EHR, providers
would still have had to select patients for whom it would be appropriate, determine
if the patients are interested and train them to use it, and then would presumably
feel responsible for checking uploaded values regularly to provide feedback. In our
center, rollout of the new function was announced across the institution but was not
accompanied by a financial or other incentive program that would have covered the
time costs of online monitoring or by an implementation plan with clinical champions
to promote use. In this naturalistic situation, providers who adopted the new function
were those who were already intensively monitoring patients. This included physicians
and nurses in the Department of Obstetrics and Gynecology, which has protocols for
routine screening for gestational diabetes and preeclampsia at 24 to 28 weeks of gestation.
For women with high-risk pregnancies, physicians promoted daily self-monitoring even
before the electronic PGHD tool was enabled. Patient-recorded values are typically
checked regularly by a physician or a nurse, and patients are encouraged to call when
values are concerning. By contrast, there is no standardized approach to PGHD data
upload for patients with chronic diabetes; therefore, uptake and workflow depend upon
provider preference. In this group, a small number of providers used the flowsheets
for patients at a time of hemoglobin A1c elevation, which may have been a time when
patients were motivated to attempt lifestyle changes or when providers were making
and monitoring changes in medical management. Others have noted that even providers
who recognize the value of PGHD still express the need for support such as practice
protocols.[3]
A second explanation for low adoption rates is that PGHD upload is likely to require
active and continued buy-in from patients. Our inclusion criteria specified that patients
must already have a patient portal account. However, it is also likely that PGHD would
be adopted only by patients who were willing to measure and input glucose values regularly
and to check provider feedback through portal messaging.[23] Self-monitoring of blood glucose can be challenging both physically and emotionally.
In some cases, patients may find self-monitoring helpful in understanding the impact
of their lifestyle and increasing their sense of control. However, in others, it may
lead to feelings of burden, frustration, anxiety, or self-blame, especially when values
are not “good” or when trends do not clearly link to lifestyle changes.[11]
[24] This is likely to reduce the numbers of patients willing to commit to routine PGHD
upload. It may also lead to abandonment after trying the tool. In many domains, people
abandon personal tracking tools after they have satisfied their curiosity, when their
health status or life circumstances change, or when tracking starts to become burdensome.[25]
[26] In other clinical domains, sustained engagement with self-monitoring has also been
linked to strong relationships with the health care provider, supportive environments,
and regular feedback.[23] In our study, another probable barrier was the need for manual data input. Although,
in theory, patients could upload glucose values directly from monitors through Apple
Healthkit, the data did not allow us to determine how many patients took advantage
of this potential convenience, and it is not clear how easy it was for them to do
this. In the future, widespread integration of PGHD into clinical care will likely
depend upon (1) targeted implementation strategies including measures to ensure that
physicians are fully incentivized to collect and review remotely collected PGHD data
rather than prioritizing face-to-face visits, (2) modifications that make technologies
easier for patients to use, such as by eliminating manual data input, (3) disease-
or population-specific decision support that uses the PGHD,[14] and (4) a stronger evidence base demonstrating ways in which the time and energy
needed to generate, review, and provide feedback on PGHD measurably improve health,
satisfaction, or other important outcomes. This will be important to ensure that patients
are helped to track information of personal value to them rather than data to be used
to meet quality measurement or value-based purchasing initiatives.
Limitations
The study is limited by its small sample size and unmeasured confounders such as health
care engagement, literacy, patient activation, and motivation, which could account
for differences between uploaders and nonuploaders. The sample size meant that it
was not feasible to apply causal inference methods such as propensity score matching.
Because of these problems, we also avoided conducting regression analysis. This makes
it impossible to rule out the possibility that the confounders (engagement with health
care and patient age) could be responsible for the observed improvement in patient
outcomes. The heterogeneity of the settings and of the patients and clinicians in
this particular dataset also makes it challenging to determine whether this technology
might be appropriate for certain subsets. Targeted studies with selected patient populations
would be needed to develop estimates of efficacy of this technology; the current analysis of this natural experiment is useful only
for showing its limited effectiveness in this specific setting. Also, this short-term analysis does not provide an evidence
on long-term effects or sustainability.
Conclusion
Despite the potential for PGHD to help medical monitoring of patients between visits,
adoption of a PGHD function within an academic multispecialty practice was slow. The
function appeared to be adopted by providers who sought to monitor patients in a period
of medical instability or transition, such as, women during high-risk pregnancy and
chronic disease patients after an elevated hemoglobin A1c measurement. Patients who
uploaded PGHD had more visits and portal uploads than patients who did not, but they
did not appear markedly sicker than nonuploaders, suggesting that they may have been
more engaged with their health care or motivated. Chronic disease patients who uploaded
PGHD data experienced improvements in their hemoglobin A1c and BMI values. However,
the improvement may have been because of motivation or increased engagement, meaning
that PGHD upload may be a support tool for patients who are already motivated rather
than one that directly improves outcomes.
Clinical Relevance Statement
Clinical Relevance Statement
Self-tracked or PGHD appears to offer the possibility of improved clinical monitoring
for patients between clinical visits and may be associated with improved outcomes.
Nevertheless, physicians may be slow to embrace PGHD data upload to the EHR unless
the process is streamlined, incentivized or reimbursed, or promoted through targeted
implementation efforts.
Multiple Choice Questions
Multiple Choice Questions
-
Diabetes patients who uploaded PGHD to a patient portal showed improvement in their
hemoglobin A1c values. Nevertheless, the effect might be because of the following:
-
Higher socioeconomic status among these patients.
-
Fewer clinical visits among these patients.
-
Higher health care utilization among these patients.
-
Improved medication safety among the patient.
Correct Answer: The correct answer is option c, higher health care utilization among these patients. Although the patients did have improved hemoglobin A1c, they also were more frequent
users of the portal and more frequent users of the health care system than nonuploaders.
This could be the explanation for their improved outcomes. By contrast, their neighborhood-level
socioeconomic status was tested and did not show any significant difference; therefore,
this is unlikely to be the explanation for the findings.
-
Patient-generated health data (PGHD) is a term that encompasses which of the following
types of information:
-
Patients answering questions about the quality of their health care.
-
Patients tracking health indicators of interest to themselves.
-
Patients describing their own health status.
-
Patient satisfaction.
Correct Answer: The best answer is option b, patients tracking health indicators of interest to themselves. The other options (a, c, and d) are instead considered patient-reported outcomes.
