Samuel Cheng
Software Engineer and UX Enthusiast
Summary
The goal of this research was to identify ways to simplify complexities within
the Explanation of Benefits (EOB) insurance document.
Even though there are problems which exist on the clinical and payer side of
financial transactions including human error in medical coding,
the solution considered in this research is empowering the end user to
deal with the confusion and frustration involved in understanding
one’s medical services. We focus specifically on a common EOB layout
by BlueCross BlueShield of Georgia. Leveraging optical character
recognition technology, we built a prototype mobile application
to present relevant medical information via an easy-to-understand interface.
Usability studies of the prototype reveal an increase in comprehension
for users and a decrease in the time it takes to achieve comprehension.
Further updates to the prototype application are forthcoming.
Details:
Details:
- August 2016 - May 2017
- Solo
- Master's Research Best Project Award: Runner-up
- User research, literature review, iterative design, Sketch, Balsamiq, Xcode, iOS, Swift, usability testing, statistical analysis
Introduction
The United States’ medical billing system is incredibly complex.
As explained in a New York Times article: “Medical bills and
explanation of benefits are undecipherable and incomprehensible even
for experts to understand” (Rosenthal, 2015).
Analysis shows that people are often confused about their medical bills and that current Explanation of Benefits (EOB) do not successfully inform patients of their coverage and charges. A 2015 survey conducted by TransUnion Healthcare found that 55% of American patients were either sometimes or always confused about their medical bills and that 61% of patients were either sometimes or always surprised about their out-of-pocket costs (TransUnion Healthcare, 2015). A 2014 Trends and Best Practices in Explanation of Benefit (EOB) Statements report conducted by DALBAR, a market research firm, found that two-thirds of EOB statements failed to help policy holders understand their medical coverage.
Confusion on medical bills can have varying consequences. More innocuously, policy holders may simply not use their health care benefits appropriately and remain ignorant about the type of services that are being performed. In addition, many patients may receive duplicate services when they visit multiple providers and specialists, often being charged unknowingly for something that was already performed. More nefariously, healthcare fraud, including medical identity theft, is one of the fastest growing crimes in the United States, costing the nation approximately $30 billion a year (Ahmed & Ahamad, 2014). By adding extra charges to a healthcare claim or coming up with completely fraudulent claims, criminals can take advantage of the patient’s confusion about the EOB report. Combating such fraud often starts by empowering and educating the patient even if the EOB is often received months after the theft has occurred (Ahmed & Ahamad, 2014).
Our goal in tackling this issue is to make the information contained in the EOB easier for patients to understand. However, this itself points to a higher goal of empowering and educating the patient to take an active role in their healthcare.
Analysis shows that people are often confused about their medical bills and that current Explanation of Benefits (EOB) do not successfully inform patients of their coverage and charges. A 2015 survey conducted by TransUnion Healthcare found that 55% of American patients were either sometimes or always confused about their medical bills and that 61% of patients were either sometimes or always surprised about their out-of-pocket costs (TransUnion Healthcare, 2015). A 2014 Trends and Best Practices in Explanation of Benefit (EOB) Statements report conducted by DALBAR, a market research firm, found that two-thirds of EOB statements failed to help policy holders understand their medical coverage.
Confusion on medical bills can have varying consequences. More innocuously, policy holders may simply not use their health care benefits appropriately and remain ignorant about the type of services that are being performed. In addition, many patients may receive duplicate services when they visit multiple providers and specialists, often being charged unknowingly for something that was already performed. More nefariously, healthcare fraud, including medical identity theft, is one of the fastest growing crimes in the United States, costing the nation approximately $30 billion a year (Ahmed & Ahamad, 2014). By adding extra charges to a healthcare claim or coming up with completely fraudulent claims, criminals can take advantage of the patient’s confusion about the EOB report. Combating such fraud often starts by empowering and educating the patient even if the EOB is often received months after the theft has occurred (Ahmed & Ahamad, 2014).
Our goal in tackling this issue is to make the information contained in the EOB easier for patients to understand. However, this itself points to a higher goal of empowering and educating the patient to take an active role in their healthcare.
Previous Work
There currently exist a few solutions targeting the visual and information layout
of the EOB report for patients. ObeoHealth is a healthcare analytics company that
provides employers with a platform for their employees to choose and manage their
healthcare options. Within the platform, their ”EOB Decoder” helps employees
understand their EOBs by presenting only the important numbers for patients to
understand along with a brief description of how charges were calculated (Rawlinson, 2015).
Though this is a good solution, it is only available to individuals whose employers
use the ObeoHealth platform.
Insurance companies such as Cigna, Humana, and Arkansas BlueCross BlueShield have taken it upon themselves to redesign their own EOBs using everyday language, easy-to-understand calculations, and better visual layout (Better “explaining” in Explanation of Benefits statements, 2016). Though this is an ideal solution, health insurance companies are under no obligation to redesign their EOBs especially due to the time and costs associated with it.
Finally, various private companies (e.g. ExperVision, ABBYY, AnyDoc EOB, and CVision) showcase their EOB imaging and capture software. These solutions allow an EOB to be scanned or faxed in upon which time the company will run an Optical Character Recognition (OCR) scan over the document and return digitized text. These OCR services work to recognize patient information and their medical costs, and analyze relationship logic of the data to confirm and check amounts (“OCR Software and Solution for Explanation of Benefits (EOBs)”). However, these solutions are only available for the provider side of the billing cycle which allows medical billers the ability to reconcile EOBs internally in their systems.
Upon review of the current solutions targeting EOBs, no patient-centered approach to ease understanding of the medical billing process has been developed.
Insurance companies such as Cigna, Humana, and Arkansas BlueCross BlueShield have taken it upon themselves to redesign their own EOBs using everyday language, easy-to-understand calculations, and better visual layout (Better “explaining” in Explanation of Benefits statements, 2016). Though this is an ideal solution, health insurance companies are under no obligation to redesign their EOBs especially due to the time and costs associated with it.
Finally, various private companies (e.g. ExperVision, ABBYY, AnyDoc EOB, and CVision) showcase their EOB imaging and capture software. These solutions allow an EOB to be scanned or faxed in upon which time the company will run an Optical Character Recognition (OCR) scan over the document and return digitized text. These OCR services work to recognize patient information and their medical costs, and analyze relationship logic of the data to confirm and check amounts (“OCR Software and Solution for Explanation of Benefits (EOBs)”). However, these solutions are only available for the provider side of the billing cycle which allows medical billers the ability to reconcile EOBs internally in their systems.
Upon review of the current solutions targeting EOBs, no patient-centered approach to ease understanding of the medical billing process has been developed.
User Research and Needs Analysis
A preliminary online survey of 21 adults of all ages (>18) was conducted to characterize
the problems that people face with EOBs. Responders were first screened to ensure that
they had basic knowledge about an EOB or had at least encountered one previously.
Corroborating the survey performed by TransUnion Healthcare, roughly 48% of responders
found EOBs to be confusing while 19% said that they were not sure.
When asked to identify specific aspects of the EOB that survey takers found confusing, the responses fell into three unique categories: text, codes, and discrepancies. Text refers to issues with font style and size, lack of visual hierarchy, density of information, and prevalence of healthcare jargon difficult for users to understand. Codes refers to the various possible codes (e.g. reason codes, CPT, LOINC) that are listed on the EOB with little to no helpful explanations. Finally, discrepancies refer to mismatches between what was estimated at the provider’s office and what the patient received on the EOB or perhaps what the patient expected their insurance to pay and what the insurance actually paid. Finally, survey takers were asked to elaborate on parts of the EOB that they would like to see improved. Some common suggestions were:
Furthermore, we must briefly mention the list of possible stakeholders in this research which include BCBSGA policy holders and their families, insurance companies, providers, and the overall health information technology (IT) community. In working to mitigate this problem, we can ultimately improve patient satisfaction and reduce costs to the healthcare industry.
When asked to identify specific aspects of the EOB that survey takers found confusing, the responses fell into three unique categories: text, codes, and discrepancies. Text refers to issues with font style and size, lack of visual hierarchy, density of information, and prevalence of healthcare jargon difficult for users to understand. Codes refers to the various possible codes (e.g. reason codes, CPT, LOINC) that are listed on the EOB with little to no helpful explanations. Finally, discrepancies refer to mismatches between what was estimated at the provider’s office and what the patient received on the EOB or perhaps what the patient expected their insurance to pay and what the insurance actually paid. Finally, survey takers were asked to elaborate on parts of the EOB that they would like to see improved. Some common suggestions were:
- ”Make the information clearer.”
- ”Translate codes into everyday language.”
- ”Include diagnostic codes.”
- ”Explain why something wasn’t paid.”
- ”Easy access to phone numbers.”
Figure 1. Sample BCBSGA EOB
Furthermore, we must briefly mention the list of possible stakeholders in this research which include BCBSGA policy holders and their families, insurance companies, providers, and the overall health information technology (IT) community. In working to mitigate this problem, we can ultimately improve patient satisfaction and reduce costs to the healthcare industry.
Initial Designs and Prototype Development
Based on the needs analysis and in light of previous work in this domain, we seek to empower
and educate policy holders by providing them an experience with their EOBs similar to that of
depositing a check on mobile banking applications. This mobile application should allow the
user to take a photo of their EOB, have an OCR engine scan and parse the text, and present clear,
concise, and useful medical information.
Design work began with pencil and paper sketches, paper prototypes, and low-fidelity prototype work in Sketch editor. Figure 2 shows a few of the screens being considered.
The application must allow the user to perform three main functions: walk them through a photo-taking and OCR scan process, view the results of the scan, and read up on the purpose of EOBs and related terminology. An initial high-fidelity prototype was developed in Xcode 8.3 using the Swift 3.0 programming language and the Tesseract OCR engine. After a few weeks of development, the resulting prototype has the following flow:
When the user first opens the application, he or she sees a "Recent Scans" page (Figure 3). Any previous scans will be shown here and future scans will be placed here as well. The user must click on the "Photos" tab in order to start the scan process. After first reading the instructions (Figure 4), the user is then presented with a screen for taking multiple pages of pictures (Figure 5 and 6). Tapping the camera icon, the user can take a photo of the EOB (Figure 7). Clicking "Use Photo" will select it and place it back into the page selection window (Figure 8). The user then taps "Process" at the top right and names the scan (Figure 9). Finally, the OCR engine will scan the text (Figure 10) and output the resulting scan when finished (Figure 11).
Design work began with pencil and paper sketches, paper prototypes, and low-fidelity prototype work in Sketch editor. Figure 2 shows a few of the screens being considered.
Figure 2. Initial design brainstorming
The application must allow the user to perform three main functions: walk them through a photo-taking and OCR scan process, view the results of the scan, and read up on the purpose of EOBs and related terminology. An initial high-fidelity prototype was developed in Xcode 8.3 using the Swift 3.0 programming language and the Tesseract OCR engine. After a few weeks of development, the resulting prototype has the following flow:
Figure 3. Recent scans tab
Figure 4. Instructions page
Figure 5. Photo page selection
Figure 6. Multi-page selection
Figure 7. Photo taken
Figure 8. Scan entered
Figure 9. Name the scan
Figure 10. Processing
Figure 11. Scan finished
When the user first opens the application, he or she sees a "Recent Scans" page (Figure 3). Any previous scans will be shown here and future scans will be placed here as well. The user must click on the "Photos" tab in order to start the scan process. After first reading the instructions (Figure 4), the user is then presented with a screen for taking multiple pages of pictures (Figure 5 and 6). Tapping the camera icon, the user can take a photo of the EOB (Figure 7). Clicking "Use Photo" will select it and place it back into the page selection window (Figure 8). The user then taps "Process" at the top right and names the scan (Figure 9). Finally, the OCR engine will scan the text (Figure 10) and output the resulting scan when finished (Figure 11).
Feedback Sessions
Three feedback sessions were conducted by members within the target user group with the
goal of assessing the usability and comprehension of the design. All three participants
were affiliated with Georgia Institute of Technology and were familiar with EOBs. Participants
were first asked to look over a paper-based EOB and give feedback about what they liked or
disliked about the medium. They were then asked to state how much they would owe based on the EOB.
Next, they were presented with a second EOB and the high-fidelity prototype application and
asked to use it to decipher the EOB data. Participants were asked to think out loud during the
session and give feedback regarding usability, design, and overall experience of using the application.
Overall, participants enjoyed the concept of the application but struggled in many usability aspects. In particular, people were unsure about the functionality of the ”Recent Scans” tab and did not know how to start a photo scan. The language of "Photos" in the middle tab was confusing in that it led people to think that they would view their photo album or at least be directly taken to a camera view instead of to an instructions page. The screen of multiple pages of pictures was particularly problematic for participants. They said that it hinted them to the check depositing features of mobile banking websites but it was still quite confusing. Participants hated having to name a scan since they did not know what to name it but instead wanted a default name supplied instead.
Overall, participants enjoyed the concept of the application but struggled in many usability aspects. In particular, people were unsure about the functionality of the ”Recent Scans” tab and did not know how to start a photo scan. The language of "Photos" in the middle tab was confusing in that it led people to think that they would view their photo album or at least be directly taken to a camera view instead of to an instructions page. The screen of multiple pages of pictures was particularly problematic for participants. They said that it hinted them to the check depositing features of mobile banking websites but it was still quite confusing. Participants hated having to name a scan since they did not know what to name it but instead wanted a default name supplied instead.
Prototype Redesign
Based off of the feedback from participants, a second iteration of the prototype was developed
using the same previous technologies. In particular, the following changes were made:
When the user first opens the application, he or she sees a "Recent Scans" page with a clear call to action (Figure 12). Tapping on the prominent camera button opens up a custom camera view with an alert dialog displayed (Figure 13). The user can swipe on the dialog to read the instructions. After closing the dialog, the user then sees the camera view with white corner brackets to help the user position the photo appropriately (Figure 14). After taking the photo, the user taps on the "Use Photo" and waits for the processing to complete (Figure 15). The OCR engine will scan the photo and output the resulting scan when finished (Figure 16). The user can now tap on the entry and view the resulting data. It includes a clear sentence or two about how much money the patient may owe as well as a small description in layman’s terms about what services were performed. A resulting pie chart shows the relationship between the full amount that was charged broken down into the amount the insurance will pay for, the amount the patient is responsible for, and the amount that is absorbed by the provider (Figure 17). In addition, the result translates the codes present on the EOB into text that that is easier for the human to understand (Figure 18) and also provides an easy-to-access number to call (Figure 19).
- The "Recent Scans" tab was redesigned to show a message when there are no existing scans and a clear call to action.
- The middle tab was redesigned to feature a more prominent camera button.
- The instructions page was removed in favor of an alert dialog.
- The multi-photo page was removed as many users would only be scanning one page.
- A custom camera view was written in order to expand the aspect ratio and add corner brackets to help guide the user when taking the photo.
- The cell heights of the scan entries were expanded to include a thumbnail of the scan and a default name.
When the user first opens the application, he or she sees a "Recent Scans" page with a clear call to action (Figure 12). Tapping on the prominent camera button opens up a custom camera view with an alert dialog displayed (Figure 13). The user can swipe on the dialog to read the instructions. After closing the dialog, the user then sees the camera view with white corner brackets to help the user position the photo appropriately (Figure 14). After taking the photo, the user taps on the "Use Photo" and waits for the processing to complete (Figure 15). The OCR engine will scan the photo and output the resulting scan when finished (Figure 16). The user can now tap on the entry and view the resulting data. It includes a clear sentence or two about how much money the patient may owe as well as a small description in layman’s terms about what services were performed. A resulting pie chart shows the relationship between the full amount that was charged broken down into the amount the insurance will pay for, the amount the patient is responsible for, and the amount that is absorbed by the provider (Figure 17). In addition, the result translates the codes present on the EOB into text that that is easier for the human to understand (Figure 18) and also provides an easy-to-access number to call (Figure 19).
Figure 12. Recent scans tab
Figure 13. Instructions alert
Figure 14. Camera view
Figure 15. Photo taken
Figure 16. Scan finished
Figure 17. Result summary
Figure 18. Services performed
Figure 19. Further help
Evaluation
After the second round of iterative design, an evaluation was necessary in order to
gauge how our target users would feel about the application and how it would affect
their comprehension of the paper-based EOBs. Our research question was established as:
Does using the application affect a user’s comprehension of information contained in the EOB?
As our hypothesis, we expected that those who use the application would exhibit greater signs of comprehension than those who use a paper EOB. A within subjects study was performed with twenty participants. 11 of the participants were male and 9 were female. 11 were faculty/staff and 9 were varying levels of students affiliated with Georgia Institute of Technology in some way. 10 were self-reported to be familiar with EOBs while the remaining 10 were unfamiliar with EOBs. The age distribution of participants consisted of 18-24 year olds: 7; 25-34 year olds: 6; 35-44 year olds: 1; 45-54: 2; 55-64: 4.
Each session of the study consisted of three main parts. First, participants were given a questionnaire to detail their familiarity with EOBs, how often they visited a doctor, and some basic demographic information. Second, participants were asked to use a paper EOB to answer a series of four questions related to comprehension and locating information as well as how confident they were in their answers. After completion of the paper EOB, they were given the prototype mobile application paired with a second EOB and asked to answer the same four questions and confidence levels. Third and lastly, they were given a post-study questionnaire for qualitative feedback and asked to rate how likely they would be to use this application in the future. The order in which the two EOB documents were presented to the participants was switched to prevent ordering effects. 9 participants performed the study using the first order of the documents while 11 participants performed the study using the other ordering. See Table 1 for more details of the study format.
The four questions that participants were asked were:
As our hypothesis, we expected that those who use the application would exhibit greater signs of comprehension than those who use a paper EOB. A within subjects study was performed with twenty participants. 11 of the participants were male and 9 were female. 11 were faculty/staff and 9 were varying levels of students affiliated with Georgia Institute of Technology in some way. 10 were self-reported to be familiar with EOBs while the remaining 10 were unfamiliar with EOBs. The age distribution of participants consisted of 18-24 year olds: 7; 25-34 year olds: 6; 35-44 year olds: 1; 45-54: 2; 55-64: 4.
Each session of the study consisted of three main parts. First, participants were given a questionnaire to detail their familiarity with EOBs, how often they visited a doctor, and some basic demographic information. Second, participants were asked to use a paper EOB to answer a series of four questions related to comprehension and locating information as well as how confident they were in their answers. After completion of the paper EOB, they were given the prototype mobile application paired with a second EOB and asked to answer the same four questions and confidence levels. Third and lastly, they were given a post-study questionnaire for qualitative feedback and asked to rate how likely they would be to use this application in the future. The order in which the two EOB documents were presented to the participants was switched to prevent ordering effects. 9 participants performed the study using the first order of the documents while 11 participants performed the study using the other ordering. See Table 1 for more details of the study format.
The four questions that participants were asked were:
- Based on the details of the EOB, how much money do you owe?
- About what percentage of your total bill will you have to pay?
- In your own words, explain what services or procedures were performed.
- What number would you call for further information or help?
Table 1. Study Format
Results
The tables below detail the results of our usability testing. For example, when asked
the first question ("How much money do you owe"), 12 out of 20 participants using the
paper EOB answered correctly with 15 being confident of their answer while 19 out of
20 participants answered correctly using the application with 19 being confident of their
answer. Only the results of the first two questions are recorded and analyzed here because
being of quantitative nature, it is easier to speak of their accuracy compared to the latter
two questions.
The results of the first question ("How much money do you owe") is analyzed here. First, an unpaired t-test was performed on the means of the confidence scores and the accuracy in order to determine whether or not the document ordering had any effect on the result. When comparing the paper EOB between documents, the resulting p-value of the confidence scores was 0.6210 and the resulting p-value of the accuracy was 0.6051 indicating that the difference in mean values between the ordering was not statistically significant. When comparing the application use between documents, the resulting p-value of the confidence scores was 0.8445 and the resulting p-value of the accuracy was 0.380 indicating that the difference in mean values between the ordering was not statistically significant. Therefore, we’ve first established that the ordering in which the two documents were presented to the participants did not have a statistically significant effect on the accuracy or confidence results.
Next, a paired t-test was performed in order to compare the means of the confidence scores and the accuracy between the use of a paper EOB versus a mobile application. The resulting p-value of the accuracy was 0.0047 and the p-value of the confidence score was 0.0002, indicating that the difference of mean values are very statistically significant.
The results of the second question ("What percentage of your total bill will you have to pay?") is analyzed next. As above, an unpaired t-test was performed on the means of the confidence scores and the accuracy in order to determine whether or not the document ordering had any effect on the result. When comparing the paper EOB between documents, the resulting p-value of the confidence scores was 0.7690 and the resulting p-value of the accuracy was 0.6051 indicating that the difference in mean values between the ordering was not statistically significant. When comparing the application use between documents, the resulting p-value of the confidence scores was 0.3193 and the resulting p-value of the accuracy was 0.8885 indicating that the difference in mean values between the ordering was not statistically significant. Therefore, we’ve first established that the ordering in which the two documents were presented to the participants did not have a statistically significant effect on the accuracy or confidence results.
Next, a paired t-test was performed in order to compare the means of the confidence scores and the accuracy between the use of a paper EOB versus a mobile application. The resulting p-value of the accuracy was 0.0102 and the p-value of the confidence score was 0.0001, indicating that the difference of mean values are very statistically significant. Digging deeper, we wanted to see whether or not other attributes of the demographics had any effect on the results. Upon closer inspection, participants who were already familiar with EOBs in some capacity performed fairly well while using the paper EOB to answer the questions. In contrast, participants who were unfamiliar with EOBs performed very poorly but their accuracy and confidence increased when using the mobile application. Table 3 and Table 4 show more detailed results of this finding.
Table 5 details the corresponding p-values. With the exception of the confidence score of the first question (which is not quite statistically significant), all the other p-values show that the differences in the means between the paper EOB and the mobile application are statistically significant.
Finally, when asked how likely they would be to use this particular application to understand their EOBs in the future, approximately 80% of the participants said that they would be highly likely to use it again.
Table 2. Overall Study Results
The results of the first question ("How much money do you owe") is analyzed here. First, an unpaired t-test was performed on the means of the confidence scores and the accuracy in order to determine whether or not the document ordering had any effect on the result. When comparing the paper EOB between documents, the resulting p-value of the confidence scores was 0.6210 and the resulting p-value of the accuracy was 0.6051 indicating that the difference in mean values between the ordering was not statistically significant. When comparing the application use between documents, the resulting p-value of the confidence scores was 0.8445 and the resulting p-value of the accuracy was 0.380 indicating that the difference in mean values between the ordering was not statistically significant. Therefore, we’ve first established that the ordering in which the two documents were presented to the participants did not have a statistically significant effect on the accuracy or confidence results.
Next, a paired t-test was performed in order to compare the means of the confidence scores and the accuracy between the use of a paper EOB versus a mobile application. The resulting p-value of the accuracy was 0.0047 and the p-value of the confidence score was 0.0002, indicating that the difference of mean values are very statistically significant.
The results of the second question ("What percentage of your total bill will you have to pay?") is analyzed next. As above, an unpaired t-test was performed on the means of the confidence scores and the accuracy in order to determine whether or not the document ordering had any effect on the result. When comparing the paper EOB between documents, the resulting p-value of the confidence scores was 0.7690 and the resulting p-value of the accuracy was 0.6051 indicating that the difference in mean values between the ordering was not statistically significant. When comparing the application use between documents, the resulting p-value of the confidence scores was 0.3193 and the resulting p-value of the accuracy was 0.8885 indicating that the difference in mean values between the ordering was not statistically significant. Therefore, we’ve first established that the ordering in which the two documents were presented to the participants did not have a statistically significant effect on the accuracy or confidence results.
Next, a paired t-test was performed in order to compare the means of the confidence scores and the accuracy between the use of a paper EOB versus a mobile application. The resulting p-value of the accuracy was 0.0102 and the p-value of the confidence score was 0.0001, indicating that the difference of mean values are very statistically significant. Digging deeper, we wanted to see whether or not other attributes of the demographics had any effect on the results. Upon closer inspection, participants who were already familiar with EOBs in some capacity performed fairly well while using the paper EOB to answer the questions. In contrast, participants who were unfamiliar with EOBs performed very poorly but their accuracy and confidence increased when using the mobile application. Table 3 and Table 4 show more detailed results of this finding.
Table 3. Paper EOB Results
Table 4. App EOB Results
Table 5 details the corresponding p-values. With the exception of the confidence score of the first question (which is not quite statistically significant), all the other p-values show that the differences in the means between the paper EOB and the mobile application are statistically significant.
Table 5. P-Values of Unfamiliar Participants
Finally, when asked how likely they would be to use this particular application to understand their EOBs in the future, approximately 80% of the participants said that they would be highly likely to use it again.
Discussion
In general, participants exhibited greater accuracy and confidence scores when using the mobile
application compared to the paper EOB. This leads us to conclude that using the mobile application
positively affects a user’s comprehension of information contained in the EOB, corroborating our
initial hypothesis. Based on the qualitative data collected from the study, participants generally
liked the clear and intuitive interface and thought that the experience of taking a photo was simple
and streamlined. One participant said that he/she "wants it now" while another stated that he/she
"didn’t have to think." The summary section with a clear bill amount was positively received and
many loved the presence of a phone number that was provided in the application itself.
However, there were some further usability issues to be addressed as well. Several users found it difficult to align the white corner brackets to the appropriate areas of the EOB, ensuring that the camera remained stable enough to take a clear photo for the OCR and then tapping the shutter button. This issue can be addressed by running algorithms to detect the scene while the camera is being aimed. The white brackets can turn green when the alignment and clarity is appropriate and automatically snapping a photo for the user. This solution also mitigates several incorrect photo-taking behaviors observed during the study. Problems included users not aligning the white corner brackets on the EOB correctly, photos not taken close enough for the text to be readable, and photos taken at incorrect angles which produced poor OCR output.
Another issue is in regards to the instructions alert dialog which we still find cumbersome to include before taking a photo. Instead, users should be taken through a paged tutorial when the application is first used which achieves the same desired goal. Finally, some users expressed that they would like to see more on the scan results page such as information about their deductible, how much has been paid to date, and even a way to manage and reconcile multiple EOBs per family member.
However, there were some further usability issues to be addressed as well. Several users found it difficult to align the white corner brackets to the appropriate areas of the EOB, ensuring that the camera remained stable enough to take a clear photo for the OCR and then tapping the shutter button. This issue can be addressed by running algorithms to detect the scene while the camera is being aimed. The white brackets can turn green when the alignment and clarity is appropriate and automatically snapping a photo for the user. This solution also mitigates several incorrect photo-taking behaviors observed during the study. Problems included users not aligning the white corner brackets on the EOB correctly, photos not taken close enough for the text to be readable, and photos taken at incorrect angles which produced poor OCR output.
Another issue is in regards to the instructions alert dialog which we still find cumbersome to include before taking a photo. Instead, users should be taken through a paged tutorial when the application is first used which achieves the same desired goal. Finally, some users expressed that they would like to see more on the scan results page such as information about their deductible, how much has been paid to date, and even a way to manage and reconcile multiple EOBs per family member.
Future Work
In addition to the issues discussed in the above section, there are several other avenues of
future work to be done in this area. First, we used the Tesseract OCR engine in order to
recognize the text and relied on post-processing in order to segment and parse the resulting text.
Further work on licensing higher quality OCR engines or development into better recognition or
post-processing algorithms should be performed. Second, as we only focused on a specific EOB
layout for BCBSGA, another step would be adding more recognition features for different types
of EOBs for different insurance companies as well.
Next, other features including the ability to adjust font size, style, and color for those with various vision impairments need to be looked into to make this mobile application accessible to those within our target population. Finally, as with any health-based application, the security and privacy of user data must be considered and protected. Currently, the application does all processing on the client-side and hence, all photos and information is stored on the phone itself. However, we do foresee the need to offload computation to a cloud service (e.g. Amazon Web Services) as more robust OCR engines require more memory for processing. We do not currently have a solution for how to manage and maintain policy holder’s privacy and security as it relates to health data but industry best practices and policies will be followed as we move forward in this area.
Next, other features including the ability to adjust font size, style, and color for those with various vision impairments need to be looked into to make this mobile application accessible to those within our target population. Finally, as with any health-based application, the security and privacy of user data must be considered and protected. Currently, the application does all processing on the client-side and hence, all photos and information is stored on the phone itself. However, we do foresee the need to offload computation to a cloud service (e.g. Amazon Web Services) as more robust OCR engines require more memory for processing. We do not currently have a solution for how to manage and maintain policy holder’s privacy and security as it relates to health data but industry best practices and policies will be followed as we move forward in this area.
References
Ahmed, M., & Ahamad, M. (2014, September). Combating Abuse of Health Data in the Age of eHealth Exchange. In Healthcare Informatics (ICHI), 2014 IEEE International Conference on (pp. 109-118). IEEE.
Better “explaining” in Explanation of Benefits statements. (n.d.). Retrieved April 13, 2016, from http://www.siegelgale.com/better-explaining-in-explanation-of-benefits-statements/
OCR Software and Solution for Explanation of Benefits (EOBs). (n.d.). Retrieved April 13, 2016, from http://www.expervision.com/find-ocr-software-by-document-types/ocr-software-for-eobs-explanation-of-benefits-1
Rawlinson, Adrian (2015, April). Can we make a better EOB? [Web log post]. Retrieved from http://www.obeohealth.com/can-we-make-a-better-eob-2/
Rosenthal, E. (2015, May 2). The Medical Bill Mystery. The New York Times. Retrieved from http://www.nytimes.com
TransUnion Healthcare. (2015, June). TransUnion Healthcare Costs Survey. Retrieved from http://transunioninsights.com/healthcarecostsurvey/
Better “explaining” in Explanation of Benefits statements. (n.d.). Retrieved April 13, 2016, from http://www.siegelgale.com/better-explaining-in-explanation-of-benefits-statements/
OCR Software and Solution for Explanation of Benefits (EOBs). (n.d.). Retrieved April 13, 2016, from http://www.expervision.com/find-ocr-software-by-document-types/ocr-software-for-eobs-explanation-of-benefits-1
Rawlinson, Adrian (2015, April). Can we make a better EOB? [Web log post]. Retrieved from http://www.obeohealth.com/can-we-make-a-better-eob-2/
Rosenthal, E. (2015, May 2). The Medical Bill Mystery. The New York Times. Retrieved from http://www.nytimes.com
TransUnion Healthcare. (2015, June). TransUnion Healthcare Costs Survey. Retrieved from http://transunioninsights.com/healthcarecostsurvey/