This is an open-access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in JMIR Medical Informatics, is properly cited. The complete bibliographic information, a link to the original publication on http://medinform.jmir.org/, as well as this copyright and license information must be included.
Cardiovascular disease causes approximately half of all deaths in patients with type 2 diabetes. Duplicative prescriptions of medication in patients with high blood pressure (hypertension), high blood sugar (hyperglycemia), and high blood lipids (hyperlipidemia) have attracted substantial attention regarding the abuse of health care resources and to implement preventive measures for such abuse. Duplicative prescriptions may occur by patients receiving redundant medications for the same condition from two or more sources such as doctors, hospitals, and multiple providers, or as a result of the patient’s wandering among hospitals.
We evaluated the degree of duplicative prescriptions in Taiwanese hospitals for outpatients with three types of medications (antihypertension, antihyperglycemia, and antihyperlipidemia), and then used an online dashboard based on mobile health (mHealth) on a map to determine whether the situation has improved in the recent 25 fiscal quarters.
Data on duplicate prescription rates of drugs for the three conditions were downloaded from the website of Taiwan’s National Health Insurance Administration (TNHIA) from the third quarter of 2010 to the third quarter of 2016. Complete data on antihypertension, antihyperglycemia, and antihyperlipidemia prescriptions were obtained from 408, 414, and 359 hospitals, respectively. We used scale quality indicators to assess the attributes of the study data, created a dashboard that can be traced using mHealth, and selected the hospital type with the best performance regarding improvement on duplicate prescriptions for the three types of drugs using the weighted scores on an online dashboard. Kendall coefficient of concordance (W) was used to evaluate whether the performance rankings were unanimous.
The data quality was found to be acceptable and showed good reliability and construct validity. The online dashboard using mHealth on Google Maps allowed for easy and clear interpretation of duplicative prescriptions regarding hospital performance using multidisciplinary functionalities, and showed significant improvement in the reduction of duplicative prescriptions among all types of hospitals. Medical centers and regional hospitals showed better performance with improvement in the three types of duplicative prescriptions compared with the district hospitals. Kendall W was 0.78, indicating that the performance rankings were not unanimous (Chi square2=4.67,
This demonstration of a dashboard using mHealth on a map can inspire using the 42 other quality indicators of the TNHIA by hospitals in the future.
Cardiovascular disease causes approximately half of all deaths in patients with type 2 diabetes [
Duplicative prescriptions refer to situations in which patients receive redundant medications for the same condition from two or more sources [
The prevalence of duplicative prescriptions is estimated at 7.4% in Japan [
From the perspective of therapeutic safety and excess expenditures, patients who receive medical care from different medical facilities are more likely to receive duplicative prescriptions and suffer adverse drug reactions [
Furthermore, increasing the transparency of hospitals is a requirement to improve administration with regard to patient safety [
By searching for the key words “duplicative prescriptions” on PubMed on April 22, 2020, only one paper [
The SNA approach [
The objectives of the present study were to (1) assess the attributes of the study data using scale quality indicators, (2) create a dashboard (ie, a control panel on a webpage that collates visual information about an issue or a topic that can be manipulated by readers themselves [
All ratio data for the three types of duplicative prescriptions on the website of TNHIA [
Descriptive statistics of hospitals included in the study.
Drug and hospital type | Taipei, n (%) | North, n (%) | Central, n (%) | South, n (%) | Kao-Pin, n (%) | East, n (%) | |
|
|
|
|
|
|
|
|
|
Medical Center (N=20) | 7 (35) | 2 (10) | 4 (20) | 3 (15) | 3 (15) | 1 (5) |
|
Regional Hospital (N=77) | 11 (14) | 17 (22) | 17 (22) | 14 (18) | 15 (19) | 3 (4) |
|
District Hospital (N=305) | 29 (10) | 62 (20) | 86 (28) | 40 (13) | 76 (25) | 12 (4) |
|
Total (N=402) | 47 (12) | 81 (20) | 107 (27) | 57 (14) | 94 (23) | 16 (4) |
|
|
|
|
|
|
|
|
|
Medical Center (N=20) | 7 (35) | 2 (10) | 4 (20) | 3 (15) | 3 (15) | 1 (5) |
|
Regional Hospital (N=79) | 11 (14) | 18 (23) | 16 (20) | 16 (20) | 15 (19) | 3 (4) |
|
District Hospital (N=308) | 29 (9) | 63 (20) | 88 (29) | 47 (15) | 69 (22) | 12 (4) |
|
Total (N=407) | 47 (12) | 83 (20) | 108 (27) | 66 (16) | 87 (21) | 16 (4) |
|
|
|
|
|
|
|
|
|
Medical Center (N=20) | 7 (35) | 2 (10) | 4 (20) | 3 (15) | 3 (15) | 1 (5) |
|
Regional Hospital (N=77) | 11 (14) | 17 (22) | 16 (21) | 16 (21) | 14 (18) | 3 (4) |
|
District Hospital (N=257) | 27 (11) | 54 (21) | 74 (29) | 31 (12) | 60 (23) | 11 (4) |
|
Total (N=354) | 45 (13) | 73 (21) | 94 (26) | 50 (14) | 77 (22) | 15 (4) |
Good data quality is necessary to ensure acceptable reliability and validity [
Therefore, before analysis, the quality of the data was assessed to ensure compliance with responses that may be producible and predictable in similar studies using the following metrics.
The reliability (ie, Cronbach α) should be greater than .70 [
The dimension coefficient [
Cronbach α tends to be overestimated. Therefore, it is recommended to rely more on convergent validity (or average variance extracted) and composite reliability values [
Where λ is the item loading to the construct domain, λ2 indicates the communality to the factor, and denotes the measurement error.
Construct reliability is also called component reliability or composite reliability, which is expressed by the following formula:
where λ and ε are defined similarly to Equation 1.
Flowchart made on a dashboard. All processes are described in detail in
The dashboard comprises the following five features: (i) the growth/share matrix of the Boston Consulting Group (BCG) on the map (ie, growth trend on the Y-axis and share on the X-axis) [
The growth (on the Y-axis, implying the trend based on recent time points) is determined by the trend via moving the control chart forward to the previous 12 months so that 24 data points yield 12 moving SDs (eg, datasets {–1,–1,–1,–1,–1,–1,–1,–1,–1,–1,–1,1} and {2,2,2,2,2,2,2,2,2,2,2,4} yield an identical correlation coefficient of 0.48 with the time series for 1 to 12), and the share (on the X-axis, indicating the accumulated momentum based on the past) is computed by the mean of the moving SDs (
Comparison of traditional control chart (top) and moving average control chart (bottom, also see
The following is a representative algorithm for locating the performance of hospitals on the four quadrants of a dashboard:
Quadrant I: the dataset {2,2,2,2,2,2,2,2,2,2,3,4} using the moving control chart forward to the previous 12 months shows continuously increasing growth (ie, y=0.63) with a positive share (ie, x=2.25).
Quadrant II: the dataset {–1,–1,–1,–1,–1,–1,–1,–1,–1,–1,1,1} shows preparedly increasing growth (ie, y=0.65) with a negative share (ie, x=–0.67).
Quadrant III: the dataset {–1,–1,–1,–1,–1,–1,–1,–1,–1,–1,–2,–3} shows good performance in controliing duplicative prescriptions with respect to growth (ie, y=–0.63) with a negative share (ie, x=–1.25).
Quadrant VI: the dataset {2,2,2,2,2,2,2,2,2,2,1,–1} indicates a decrease in growth (ie, y=–0.60) when the share is still positive (ie, x=1.67).
We used the analytic hierarchical process [
Calculation of weights for evaluating and ranking hospital performance. In step 1, scores are assigned from 3 (best, green) to 1 (worst, red). In step 2, pair comparison (eg, 3/2=1.5, 2/1=2, 1/3=0.3, etc) is performed to obtain the odds for each cell in the top panel. In step 3, the odds/summation ratio is calculated for each cell in the bottom panel, and the bottom row is averaged to obtain the final weight (eg, 0.5, 0.3, and 0.2).
Finally, we used Kendall coefficient of concordance (
SPSS 19.0 for Windows (SPSS Inc, Chicago, IL, USA) and MedCalc 9.5.0.0 for Windows (MedCalc Software, Mariakerke, Belgium) were used to calculate Cronbach α, dimension coefficients, and other scale quality indicators used in this study. The cloud computation was programmed using the active server pages on the website (see
The scaling quality for the study data was found to be acceptable (dimension coefficient>0.67 and Cronbach α>.70), indicating that these duplicative prescription ratio data are reliable and consistent with our expectation (
Quality assessment of the study data.
Type of duplicative prescription | Dimension coefficient | Cronbach α (reliability) | Average variance extracted | Construct reliability |
Antihypertension | 0.69 | .79 | 0.80 | 0.99 |
Antihyperglycemia | 0.73 | .91 | 0.85 | 0.99 |
Antihyperlipidemia | 0.71 | .88 | 0.75 | 0.98 |
The dashboards shown in
Dashboard of antihypertension duplicate prescription performance.
Dashboard of antihyperglycemia duplicate prescription performance.
Dashboard of antihyperlipidemia duplicate prescription performance.
As shown in
Kendall
Frequency of the three types of duplicative prescriptions in the four quadrants on the dashboards.
Prescription and hospital type | Red (weight=0.2), n (%) | Yellow (weight=0.3), n (%) | Green (weight=0.5), n (%) | N | Score | Chi square (df=4) | ||||||||
|
|
|
|
|
|
|
|
|||||||
|
Medical Center | N/Aa | N/A | 20 (100) | 20 | 50.0 |
|
|
||||||
|
Regional Hospital | N/A | 1 (1) | 76 (99) | 77 | 49.8 |
|
|
||||||
|
District Hospital | 8 (2) | 170 (56) | 127 (42) | 305 | 38.2b |
|
|
||||||
|
Total | 8 (2) | 171 (42) | 223 (56) | 402 | N/A |
|
|
||||||
|
|
|
|
|
|
|
||||||||
|
Medical Center | N/A | 1 (5) | 19 (95) | 20 | 49.0 |
|
|
||||||
|
Regional Hospital | N/A | 6 (8) | 73 (92) | 79 | 48.4 |
|
|
||||||
|
District Hospital | 13 (4) | 156 (51) | 139 (45) | 308 | 38.6 |
|
|
||||||
|
Total | 13 (3) | 163 (41) | 231 (56) | 407 | N/A |
|
|
||||||
|
|
|
|
|
|
|
|
|||||||
|
Medical Center | N/A | 1 (5) | 19 (95) | 20 | 49 |
|
|
||||||
|
Regional Hospital | N/A | 2 (3) | 75 (97) | 77 | 49.4 |
|
|
||||||
|
District Hospital | 13 (5) | 143 (56) | 101 (39) | 257 | 37.3 |
|
|
||||||
|
Total | 13 (4) | 146 (41) | 195 (55) | 354 | N/A |
|
|
aN/A: not applicable.
bScore is calculated as: 38.2=(2%×0.2+56%×0.3+42%×0.5)×100.
Rankings of hospital type for duplicative prescriptions.
Hospital type | Antihypertension | Antihyperglycemia | Antihyperlipidemia |
Medical center | 1 | 1 | 2 |
Regional hospital | 2 | 2 | 1 |
District hospital | 3 | 3 | 3 |
We used dashboards with an mHealth tool to create an animated dashboard that represents the hospital performance sheet of managing duplicative prescriptions in Taiwan. The data quality were acceptable and effectively reflected the reliability and construct validity. The online dashboards enabled easy and clear interpretation of duplicative prescriptions related to hospital performance using multidisciplinary functionalities, demonstrating a trend toward reducing duplicative prescriptions among all types of hospitals. Medical centers and regional hospitals exhibited better performance improvement for reducing duplicative prescriptions for the three types of controlled medications compared with district hospitals. Kendall
Many researchers have published studies based on Google Maps [
Making hospitals more transparent [
We also found that many district hospitals have incomplete (or missing) data on the ratio of duplicative prescriptions. The reason might be that many district hospitals are significantly affected by the global budget payment system, forcing them to terminate their businesses due to difficult operations in health services.
Management differentiation strategies [
The use of weights that should sum to 1.0 (as illustrated in
Google Maps provides programmers with an API to incorporate coordinates with visual representations and build a dashboard-type report card. We demonstrated the process of creating HTML in the video of
The TNHIA website [
As mentioned above, the data quality should be ensured before analysis. This task involves examining the responses that are consistent and reproducible with acceptable reliability and validity [
We evaluated the scale quality with several indicators based on classical test theory. Furthermore, we illustrated the importance of the API in
Several issues should be considered thoroughly in the future. First, the study data were incomplete, especially for the district hospitals. Thus, inference making, such as for district hospitals with poor performance in controlling duplicative prescriptions, should be conservative. This limitation calls for further research and validation.
Many innovations have been introduced with advances in science and technology, such as the visual dashboard on Google Maps using the coordinates to display and line plots on cloud computation as shown in
Third, the mascots illustrated in the BCG matrix, such as stars, problem children, cash cows, and dogs, might be inappropriate in health care settings. Other mascots such as Santa Claus, productive cows, or dejected dogs, could refer to appropriate dashboard-type report cards in the future.
Fourth, the scaling quality for the study data was found to be acceptable (ie, dimension coefficient>0.67 and Cronbach α>.70), indicating that these duplicative prescription ratio data are reliable and consistent with our expectation. The dimension coefficients were relatively low (ie, 0.69, 0.71, and 0.73), indicating that all datasets were weak when measuring a one-dimensional feature (ie, duplicative prescriptions). Therefore, there is low confidence when using the result to make an inference for the future. Further studies should pay more attention to the issue of data fitting to the unidimensional requirement.
Fifth, the effect of weights was obvious due to different sample sizes in different hospital types. We normalized the summed weights to be 1.0 and ensured fair comparisons among hospital types across performance categories (ie, red, yellow, and green bubbles). If the percentages of the performance categories differ among hospital types, the weights will affect the assessment results. For this reason, we used an analytic hierarchical process [
This study provides a demonstrated platform with an online quality report card on detecting the performance of duplicative prescriptions to help health care practitioners easily upload data and quickly provide feedback on visual representations on an online dashboard. These dashboards can be used to build an online report card for hospitals under supervision of the public based on mHealth and uHealth in the future.
MP3: How to build Google maps for this study.
The moving average control chart used in this study.
Excel dataset.
MP3: How to manipulate the mHealth dashboard on Google Map.
application programming interface
Anatomical Therapeutic Chemical
Boston Consulting Group
hypertext markup language
mobile health
social network analysis
Taiwan National Health Insurance Administration
WC and SC developed the study concept and design, and drafted the manuscript. SC, JU, and YT analyzed and interpreted the data. PH monitored the process of this study. All authors provided critical revisions for important intellectual content. The study was supervised by TW. All authors read and approved the final manuscript.
None declared.