Abstract
When assessing the characteristics and performance of telemedicine interventions, most studies followed a patient- centric approach, leaving the telemedicine providers’ role out of consideration. As a result, little was known about the demographics and prescription pattern of telemedicine physicians, the knowledge of which is integral to a holistic evaluation of the virtual delivery of accountable care. To fill this gap, our study explored how physicians’ traits and encounter-specific characteristics correlate with prescription outcomes, using multivariate analyses. Significant inter-physician variation in prescription behaviors was observed and analyzed in sub-groups. The average Virtual Urgent Care physician’s prescription likelihood was 69% with a mean prescription count of 0.98; male physicians and primary care providers tended to prescribe both more often and with a greater number of medications. This study called attention to the quality and reproducibility of telemedicine providers’ prescription decision and warned the likely absence of well-defined practice guidelines for delivering virtual care.
Introduction
Telemedicine facilitates patient-provider interaction and the delivery of clinical consultation via information technology1. While most telemedicine research focuses primarily on patient’s characteristics, clinical effectiveness, and technical outcomes2-5, limited number of studies investigated the physicians’ prescribing behaviors in a virtual setting.
Telemedicine Providers’ Prescribing Patterns
In the past, researchers were primarily concerned with telemedicine providers’ performance in terms of diagnostic accuracy, obtained by analyzing medical images in certain specialties such as radiology, dermatology, ultrasound, and pathology6. However, this approach was not applicable to common acute illnesses, which made up the majority of current commercial virtual visits7. Only a handful of studies shed light on the demographics and practicing behaviors of telemedicine providers. Previously, researchers studied the difference between in-person and telemedicine providers’ prescribing patterns, concluding that provider type was a significant influence in determining the count of medication prescribed per consultation8. Another study investigated the prescription behaviors of General Practitioners in a Danish out-of-hours primary care telephone consultation practice. The authors further explored the association between various patients and physicians’ characteristics and prescription frequency, concluding that one in five phone consultations ended with a prescription; the prescription likelihood was the highest from 4 to 8 p.m. on weekdays and during the early morning on weekends9.
These findings, to some extent, complemented the previous research on inter-physician variation in prescribing practice in the traditional, face-to-face setting, which generally suggested a positive correlation between the providers’ personal characteristics and prescription outcomes10-12. However, the presence of a significant knowledge gap in telemedicine physicians precluded a systematic, direct comparison between the in-person and virtual physician populations. Therefore, the knowledge of telemedicine providers will offer valuable insights on the patient-physician relationship in the virtual environment, which has become an urgent matter as telemedicine gained more acceptance in recent years.
Virtual Urgent Care
This study is uniquely important in a sense that it investigated a rarely-studied type of telemedicine service—a Virtual Urgent Clinic (VUC) telemedicine platform, providing virtual medical consultation to the general population in North Carolina 24-hours a day, 7-day a week. More importantly, the VUC physicians come from four areas of specialties, which enabled between-provider-type comparison. Encounters characteristics, such as time of day, day of the week, modality, and encounter duration were analyzed to provide a comprehensive understanding of telemedicine providers’ prescribing behaviors.
Objective
The goals of this paper were to further understand the characteristics of VUC physicians and to identify pattern and variation in their prescribing behaviors, using a combined method of descriptive and multivariate analysis. In addition, we discussed the implications of the providers’ behaviors in improving the delivery of accessible and accountable virtual care.
Methods and Materials
Program Description
We investigated the first 11-month operation data from the 24/7 virtual urgent care service in 2018. The VUC provides patient care across North Carolina with synchronous, virtual consultation for illnesses of acute nature—around-the- clock and 7 days a week—enabled by a third-party telemedicine platform.
Registration is mandatory for all patients in order to access the VUC service. Upon registration, the users are required to provide both demographic data (including age, gender, zip codes of residence, and status of dependency) and detailed accounts of medical history such as allergies, drug sensitivities, and family health history. At the point of scheduling a consultation, the patients are presented with a comprehensive list of conditions which could not be treated via the platform. The patients specify the reason for the visit from a pre-identified list of conditions and choose a provider per one’s preferences and availability. Moreover, the patient will choose either a phone call or video conference as their preferred encounter medium. Additional instructions are provided in case where the required software has not been appropriately installed. During the consultation, the VUC provider evaluates the patient symptoms and document the diagnosis in the standard ICD-9-CM Diagnosis Code; in situations which result in a prescription of any medications, the physician will send an electronic prescription to the patient’s pre-assigned pharmacy. The VUC service cost is a flat fee regardless of consultation types or duration. The study was approved by the Institutional Review Board at the University of North Carolina-Chapel Hill.
Study Population
VUC delivers urgent medical care to residents of North Carolina above the age of two-year-old. In addition, the patients will need access to a smartphone or a computer with the required software installed. VUC serves patients regardless of their demographic traits, insurance coverage, or other status.
All VUC third-party providers were board-certified physicians. We collected data of 26 VUC providers, including 9 Family Medicine, 8 Internal Medicine, 6 Emergency Medicine, 2 Pediatrics, and 1 General Practice physician. The sole General Practice provider was excluded from the analysis due to its inability to represent the General Practice physician population, which is likely to compromise the overall quality of the study. We decided to focus on two main types of providers: Primary Care—which encompassed Family Medicine, Internal Medicine, and Pediatrics providers14—and Emergency Medicine.
Dataset and Variables
For analysis purpose, 141 incomplete and test-user encounters were excluded. Eight independent variables from two categories, the physicians’ characteristics and the encounter-specific features, were extracted and coded.
The provider’s gender, enunciation date, and medical specialty were identified through looking up their National Provider Identifiers(NPI), which was included in the original dataset, on the NPPES NPI Registry website. The procedure was aided by applying generic web-scraping algorithm in Python. The variable Year of Practice was constructed by calculating the difference between the physician’s year of enunciation and the current year (2019) and was subsequently categorized into over 10 years and under 10 years. Additionally, we counted the number of encounters completed by each provider and constructed Practice Frequency, indicating an individual’s degree of involvement in practicing telemedicine; likewise, the variable was divided into three categories—over 100 times, 51- 100 times, and 11-50 times. Encounter-specific features were also recorded in the original VUC dataset, including Time of Day, Day of Week, Encounter Modality, and Encounter Duration. The features were classified into subgroups for the purpose of evaluation, as shown in Table 1. For additional inter-physician prescription variation analysis, we divided the dataset by the top five popular diagnosis codes: Sinusitis, Urinary Tract Infection, Pharyngitis, and Bronchitis).
Table 1.
Average Prescription Likelihood and Prescription Count by Physicians and Encounter characteristics.
| Population N (%) | Presc.likelihood | Presc. likeli hood 95% CI | Presc. Count | Presc. Count 95% CI | ||
| Physicians Characteristics | All Encounters (N) | 1,217(100%) | 0.69 | 0.03 | 0.98 | 0.05 |
| Genderαβ | ||||||
| Female | 347(28.51%) | 0.52 | 0.05 | 0.73 | 0.09 | |
| Male | 870(71.49%) | 0.76 | 0.03 | 1.08 | 0.06 | |
| Year of Practiceα | ||||||
| Over 10 years | 1031 (84.72%) | 0.68 | 0.03 | 0.92 | 0.05 | |
| Under 10 years | 186 (15.28%) | 0.72 | 0.07 | 1.34 | 0.17 | |
| Specialtyαβ | ||||||
| Emergency Medicine | 384(31.55%) | 0.64 | 0.05 | 0.86 | 0.08 | |
| Primary Care | 833(68.45%) | 0.71 | 0.03 | 1.04 | 0.06 | |
| Prescription Frequencyαβ | ||||||
| 11-50 times | 496(40.76%) | 0.71 | 0.04 | 0.98 | 0.07 | |
| 51-100 Times | 269(22.1%) | 0.56 | 0.06 | 0.95 | 0.13 | |
| Over 100 Times | 452(37.14%) | 0.75 | 0.04 | 1.00 | 0.08 | |
| Encounter Characteristics | ||||||
| Time of Dayβ | ||||||
| 6 a.m. - 12 p.m. | 484(39.77%) | 0.73 | 0.04 | 1.04 | 0.08 | |
| 12 p.m. - 5 p.m. | 364(29.91%) | 0.65 | 0.05 | 0.93 | 0.10 | |
| 5 p.m. - 10 p.m. | 272(22.35%) | 0.69 | 0.06 | 0.99 | 0.11 | |
| 10 p.m. - 6 a.m. | 97(7.97%) | 0.61 | 0.10 | 0.86 | 0.18 | |
| Day of weekαβ | ||||||
| Weekday | 896(73.62%) | 0.71 | 0.03 | 1.00 | 0.06 | |
| Weekend | 321(26.38%) | 0.63 | 0.05 | 0.92 | 0.11 | |
| Duration of encounter | ||||||
| <1 Min | 21(1.73%) | 0.76 | 0.20 | 0.95 | 0.34 | |
| 1-5 Min | 839(68.94) | 0.70 | 0.03 | 1.00 | 0.06 | |
| 6-10 Min | 289(23.75) | 0.64 | 0.06 | 0.92 | 0.10 | |
| >10 Min | 70(5.59) | 0.69 | 0.11 | 1.01 | 0.22 | |
| Model | ||||||
| Phone | 1071(88) | 0.69 | 0.03 | 0.99 | 0.05 | |
| Video | 146(12) | 0.68 | 0.08 | 0.93 | 0.14 |
*α denotes Kruskal-Wallis rank sum test p-value <0.05 for characteristics—prescription count.
**β denotes Chi-squared test p-value <0.05 for characteristics –prescription likelihood.
Outcome Variables
The primary outcomes of interest were prescription likelihood and prescription count. Prescription Likelihood denoted the possibility of whether an encounter ended with medication prescribed; it was calculated by averaging the binary prescription outcome of encounters under a certain criterion, where 0 indicated no prescription and 1 indicated at least one count of medication. The variable Prescription Count recorded the exact number of medications prescribed per session on a range of 0 to 6.
Statistical Analysis
Descriptive analysis was conducted on the test population, giving the mean prescription likelihood and count by physicians and encounter characteristics. For the count variable Prescription Count, we applied the Kruskal-Wallis test to determine if the means in a group were significantly different from one another. The non-parametric Kruskal- Wallis test was chosen over the One-way ANOVA since the former does not assume the normality condition13. For the nominal variable Prescription Likelihood, Pearson Chi-squared test was used for assessing the in-group difference. In addition, the top five most popular diagnoses were extracted and tested individually for the presence of inter-physician difference in prescribing outcome using Chi-squared test.
Variable Selection and Modeling
To remove the unnecessary predictors, a Stepwise Regression was run with all eight independent variables. Five predictors were kept for the final modelling while four were excluded. Doing so help reducing the noise which may arise from the collinearity among the redundant predictors.
Subsequently, a Multivariate Logistic Regression model was constructed to investigate the association between Prescription Likelihood and the five independent variables; log ratios and p-values were calculated. In addition, a Poisson Regression was built for estimating the impact of the independent variables on Prescription Count. Based on Cameron and Trivedi’s recommendation in their 2009 study, we calculated the robust standard errors for the parameter estimates in order to compensate for mild violation of the equal variance assumption14.
Data Collection and Cleaning
Encounter data was automatically scrapped from the service’s website and stored in a secure SAP© Business Objects (BO) enterprise system. The dataset included physicians’ unique identifier, diagnosis code, prescription count, and encounters characteristics such as time of day, day of the week, duration, and modality information. The dataset was cleaned and preprocessed with Excel and OpenRefine. The statistical analyses and models were generated using R (R Foundation for Statistical Computing, 2014). The data visualization was made with Tableau Desktop 2019.
Results
This study analyzed a total of 1,217 encounters completed by 25 physicians, collected during the 11-month period between February 8th and November 29th, 2018. On average, 69% of VUC encounters ended with at least one medication prescribed, with a mean prescription count of 0.98 per encounter.
Inter-physician Variation in Prescribing Behavior
To evaluate the presence of inter-physician variation in prescribing behavior, we assigned each provider a unique identifier, Provider ID (ranging from 1 to 25), which was subsequently tested as a predictor of prescription outcome. A Pearson Chi-squared test of Provider ID and Prescription Likelihood yielded a p-value of 0, indicating an extremely strong correlation between the individual provider’s accumulated characteristics and prescription likelihood. In other words, the 25 physicians prescribed in significantly different manners. Figure 1. visualized the frequency of prescribing a certain number of medications (0, 1, 2, and 3+) as a percentage of a provider’s total encounters.
In encounters resulting in prescription, one count of medication was the most common case. However, some physicians showed a tendency of prescribing multiple medications per encounter; for instance, provider 6 prescribed more than one medication per session about 90% of the time and frequently prescribed greater than three counts. On the other hand, some providers showed reluctance towards prescribing at all (Provider 4, 12, 23), Figure 1.
Figure 1.
Individual Difference among VUC Physicians’ Prescription Counts.
Descriptive Analysis
Providers Characteristics
Among the 1,217 encounters, 71.5% were completed by male physicians, who, on average, prescribed 24% more frequent than the female counterpart, Table 1. Of all consultations, 84.72% were completed by providers with over ten years of experience. Physicians with less of ten years of practice tended to prescribe a significantly greater number of drugs per encounter; however, they were not more likely to prescribe than their counterpart. Furthermore, Primary Care providers completed the most number of consultations (68.45%), while Emergency Medicine contributed the remaining 31.55%. The Chi-squared test results of both outcome metrics indicated significant difference between provider types. Lastly, 40.76% of the consultations were conducted by physicians who practiced less often (11-50 times), while physicians who had completed over 100 sessions contributed 37.14%. The between-group differences were significant in both outcome measures.
Encounters Characteristics
Of all encounters, 39.77% took place in the morning (6 a.m.- 12 p.m.); afternoon (12 p.m.- 5 p.m.) and evening sessions (5 p.m.- 10 p.m. and 10 p.m.-6 a.m. combined) composited around 30% of total encounters respectively. Physicians were more likely to prescribe during morning hours, where 73% of sessions ended with drugs prescribed; sessions happened after 10 p.m. had the lowest prescription likelihood of 61%. Likewise, morning sessions had the highest prescription count (1.04), which is 73.56% higher than the average count of late night sessions (0.84). Additionally, approximately 7 out of 10 encounters lasted between 1 and 5 minutes. Durations of less than 1minute and over 10 minutes were rare, with less than 2% and 6% rates of occurrence respectively. Furthermore, about 3 out of 10 encounters happened during the weekend, the prescription likelihood and count of which were both significantly lower than those of the Weekday encounters. In terms of telemedicine modality, Phone (88%) was predominantly more popular than Video (12%); the difference in prescription outcomes between the two, however, was not significant.
Prescription Outcomes Analysis
Prescription Rate
To further understand whether a certain physician or encounter characteristic could potentially affect the likelihood of prescription at a given consultation encounter, a Multivariate Logistic Regression was run to determine both the significance and magnitude of each independent variable’s predictive power, as shown in Table 2.
Table 2.
Multivariate Logistic Regression and Poisson Regression.
| Dependent Var. | |||||
| Prescription Rate | Prescription Count | ||||
| Independent Var. | Odds Ratio | p value | Incidence Rate Estimate | p value | |
| (Intercept) | 0.78 | 0.36 | 0.58*** | 0 | |
| Time of Day | 12 p.m. -5 p.m. | 1.04 | 0.86 | 1.09 | 0.49 |
| 5 p.m. -10 p.m. | 1.31 | 0.29 | 1.15 | 0.23 | |
| 6 a.m.-12 p.m. | 1.59* | 0.06 | 1.26 | 0.05 | |
| Day of Week | Weekend | 0.78* | 0.08 | 0.96 | 0.51 |
| Gender | Male | 3.10*** | 0.00 | 1.48*** | 0 |
| Specialty | Primary Care | 1.42** | 0.02 | 1.15** | 0.02 |
| Practice Frequency | 51-100 times | 0.48 | 0.38 | 0.97 | 0.68 |
| Over 100 times | 1.15 | 0.40 | 0.99 | 0.87 | |
Akaike Inf. Crit. 1,355.739
Null deviance: 1150.3 on 1216 degrees of freedom
Residual deviance: 1103.1 on 1208 degrees of freedom
AIC: 2991.5
Note: *p<0.1; **p<0.05; ***p<0.01
The resulting test statistics indicated that Provider Gender and Specialty was the most significant predictors, with p values of less than 0.05; Time of Day and Day of Week were less strong but still marginally significant, with p-values of less than 0.1. Odds ratio were subsequently calculated. For instance, all else held equal, male providers were 2.1 times more likely to prescribe than their female counterpart. Primary Care providers were 42% more likely to prescribe than the EM specialists. On the other hand, a Weekend encounter, completed by a physician with the same set of characteristics and shared the identical encounter features as its Weekday counterpart, was 78% less likely to end with medication prescribed. In addition, consultations happened in the morning hours were about twice more likely to get prescription as compared to the sessions taken place after 10 p.m..
Prescription Count
A Poisson Regression with Prescription Count and the set of 5 variables was performed. To ease the interpretation, we transformed the resulting log-odds estimate into incidence rate ratio, Table 2.
Provider Gender and Specialty turned out to be significant predictors, with p-values of less than 0.01 and 0.1 respectively. All else features held equal, male physicians prescribe 48% unit more than the female. Under the same encounter settings, Primary Care providers prescribe 15% more than the EM specialists of their own gender and practice frequency.
Prescription Outcome Analysis for the Top 5 Diagnosis
The five most popular diagnosis (accumulated n=463, 38.04% of total encounters) were individually tested for inter- physician variation in prescribing outcome. The statistics were obtained using Chi-squared test, as shown in Table 3.
Table 3.
Prescribing Patterns in the Top 5 Diagnosis
| Diagnosis Name (ICD-10-CM Code) | Count | Presc. Likelihood | Presc. Likelihood 95% CI | p-value1 |
| Sinusitis (J01.90) | 198 | 0.833 | 0.781-0.886 | 0*** |
| Urinary Tract Infection (N39.0) | 86 | 0.80 | 0.716-0.888 | 0.003*** |
| Pharyngitis (J02.0) | 65 | 0.708 | 0.594-0.821 | 0.02** |
| Upper Respiratory Infection | 58 | 0.638 | 0.510-0.765 | 0.05** |
| (J06.9) | ||||
| Bronchitis (J20) | 56 | 0.857 | 0.763-0.952 | 0.37 |
| Total | 463 | 0.788 | 0.751-0.826 | 0 |
Note: p-value was calculated using Chi-squared Test.
*p<0.1; **p<0.05; ***p<0.01
Different physicians prescribed in significantly different manner when treating Sinusitis, Urinary Tract Infection, Pharyngitis, and Upper Respiratory Infection (p value<0.05). However, the inter-physician variation was not observed in treating Bronchitis. Bronchitis patients were also the most likely to get prescription (prescription likelihood =85.7%). In case of Upper Respiratory Infection, only 63.8% encounters ended with any drug prescribed.
Discussion
Main Findings
We found that virtual care providers were predominantly male primary care physicians with over ten years of experience. Male providers displayed a tendency toward prescribing a greater number of medications at any given consultation. In addition, the difference between Primary Care and Emergency Medicine physicians were evident in both outcome measures, with the latter being more reluctant toward prescribing and suggesting a higher count of drugs. A marginally significant correlation existed between temporal factors (Time of Day, Day of the Week) and prescription outcomes. On the other hand, the impact of telemedicine modality and duration of encounter were not evident in affecting the physician’s prescribing behaviors. Inter-physician variation in prescription outcome was the most present in treating diseases of the respiratory system (ICD-10-CM Diagnosis Code J00-J99), which were also the most popular VUC diagnoses.
Previous Study
Telemedicine Interventions
In terms of prescription outcomes, the variations were considerable: The average prescription likelihood of VUC encounters was 0.69, which was significantly higher than the rate of 0.19 in a similar study of a telephone consultation intervention in Denmark9. On the other hand, VUC’s mean prescription count was 0.98, which was 29.6% lower than the 1.27 average in another e-visits intervention8. However, the scarcity of comparable literature prevented us from understanding the underlying pattern of such pronounced differences among individual telemedicine interventions. These great variances alarmed the potential absence of common practice guidelines in the virtual setting, raising questions on the fairness and quality of telemedicine delivery.
In-Person Urgent Care Visit
According to a national survey conducted with 436 in-person Urgent Care Centers in the United States, primary care providers managed 72% of total encounters, while emergency medicine physicians and non-primary care specialists contributed 28% and 20% respectively17. Our study indicated a similar composition of physician’s specialty, with a 2:1 ratio between primary care and emergency medicine physicians. However, there was an absence of non-primary care specialists among the telemedicine providers. We suspected that the VUC protocol, designed specifically for the assessment of common acute illnesses via voice or video conferencing, prohibited advanced patient-physician interaction that might be required in the specialists’ consultation.
Time and day of urgent care visit were significantly different between in-person and virtual settings. One study sampled 387,746 records of acute care visit in the U.S., collected between 1997 and 2010, suggesting that 95 percent of visits to office-based primary care providers happened on weekdays18. On the other hand, three out of ten VUC encounters took place on weekends. The virtual urgent care demonstrated potential in providing more timely and accessible medical service due to the around-the-clock availability of telemedicine physicians. In other words, telemedicine service could complement the office-based delivery of primary care, ameliorating the staffing problem faced by many healthcare organizations nowadays19.
In-person Physician
Our study agreed with a number of researches on that male physicians tend to be high-prescribers10,11. One study concluded that physicians who had more practice days and saw more patients per day would also prescribe greater number of drugs11. While our findings indicated significant between-group differences among telemedicine physicians with varied frequency of practice, the variable itself was not an effective predictor of both prescribing outcomes in a virtual setting.
Strengths and Limitations
This study was the first to investigate the characteristics and prescription behaviors of virtual urgent care telemedicine providers, with first-hand data collected from a Southeastern commercial telemedicine platform. In addition, we searched for literature on office-based urgent care centers in the United States and compared the operational statistics with the VUC dataset. However, the inter-physician variation in prescribing styles suggested that one’s prescribing decision might be prone to personal biases, which we were not able to validate in this paper.
In addition, this paper did not directly compare telemedicine physicians with in-person physician due to the lack of access to comparable data collected from in-person urgent care settings. We were only able to compare our findings with similar telemedicine evaluation studies. The implications of this paper could be further strengthened if comparison was made between virtual and traditional urgent care providers.
On the other hand, we only studied the five most popular diagnostic groups, which made up 38.04% of total encounters, instead of analyzing all diagnose. The reason for this is that the original dataset came with 124 types of diagnoses, which we were unable to categorized into meaningful groups without expert knowledge in disease classification. Therefore, further studies may focus on analyzing a selective group of diagnoses, which would yield more disease- specific findings.
Last but not least, the quality and reproducibility of telemedicine consultation remained in question. To our knowledge, only one study evaluated the reproducibility of telemedicine diagnosis of common acute problem, suggesting a strong trend for telemedicine physicians to disagree on primary diagnosis made by office-based physicians20. Future work could test the inter-physician agreement on prescription outcome of a certain type of acute illness. Additional test could be designed to assess the unbiasedness of telemedicine provider, measured by the consistency of prescription outcome generated by an individual provider.
Future Direction
Previous studies highlighted the pivotal role of an extensive examination of both the quantitative and the qualitative determinants in order to understand the motivation and preference behind the physician’s prescribing behavior21,22. Therefore, future research could incorporate qualitative metrics that measure personal, organizational, and sociology- technical qualities: the physicians’ attitudes toward telemedicine, work ethics, and organizational culture, etc. Such measures could be obtained through survey questions or contextual interview.
On the other hand, traditional urgent care and ambulatory care had been known for antibiotics overuse. According to one study, about 1 in 3 antibiotics prescribed at ambulatory care visits was unnecessary23. Researchers could analyze the antibiotics prescription pattern in the telemedicine interventions, which will add to the current knowledge of antibiotic abuse.
Moreover, it was unknown whether the virtual care providers had received telemedicine-specific training prior to practicing online. We highly recommend telemedicine developers and healthcare organization leadership to consider developing training and educational programs in order to standardize and regulate prescription appropriateness in the virtual setting.
Conclusion
In conclusion, this study investigated physicians prescribing behaviors in a telemedicine urgent care platform. We identified that male primary care physicians made up the majority of the telemedicine provider population; this subgroup was also the most likely to prescribe a greater number of drugs at a given consultation. Primary Care providers and EM physicians showed different prescribing styles, with the latter appeared to be more reluctant toward prescribing online. However, this study alone could not explain the inter-physician variation in prescribing behavior due to the lack of access to qualitative measures such as the physician’s attitude toward practicing online.
Nevertheless, this study highlighted the difference in prescribing pattern among telemedicine providers, which could potentially compromise the quality of virtual care delivery. We urged future researchers to focus on the appropriateness and reproducibility of telemedicine physician’s prescribing outcomes.
While telemedicine has been increasingly perceived as a timely, cost-effective alternative to traditional office-based medical care delivery, it is crucial to ensure that the quality of care is not compromised. We believe that an enhanced understating of telemedicine providers’ behaviors would help improving the delivery of around-the-clock, appropriate, and accountable medical care to the population.
Figures & Table
References
- 1.Perednia DA, Allen A. Telemedicine Technology and Clinical Applications. J Am Med Assoc. 1995;273((6)) [PubMed] [Google Scholar]
- 2.Jung C, Padman R. Virtualized healthcare delivery: Understanding users and their usage patterns of online medical consultations. Int J Med Inform. [Internet]. 2014;83((12)):901–14. doi: 10.1016/j.ijmedinf.2014.08.004. Available from: [DOI] [PubMed] [Google Scholar]
- 3.Martich GD. Internet-Based Medical Visit and Diagnosis for Common Medical Problems: Experience of First User Cohort. Telemed e-HEALTH. 2010;17((4)):304–8. doi: 10.1089/tmj.2010.0156. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Mehrotra A, Paone S, Martich GD, Albert SM, Shevchik GJ. Characteristics of Patients Who Seek Care via eVisits Instead of Office Visits. Telemed e-HEALTH. 2013;19((7)):515–9. doi: 10.1089/tmj.2012.0221. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Hersh WR, Hickam DH, Severance SM, Dana TL. Diagnosis, access, and outcomes: update of a systematic review of telemedicine services William. J Telemed Telecare. 2006;12((Supplement 2)):3–31. doi: 10.1258/135763306778393117. [DOI] [PubMed] [Google Scholar]
- 6.Aoki N, Dunn KIM, Johnson-Throop KA, Turley JP. Outcomes and Methods in Telemedicine Evaluation. Telemed e-Health. 2003;9((4)):393–401. doi: 10.1089/153056203772744734. [DOI] [PubMed] [Google Scholar]
- 7.Schoenfeld AJ, Davies JM, Marafino BJ, Dean M, Dejong C, Bardach NS, et al. Variation in Quality of Urgent Health Care Provided During Commercial Virtual Visits. JAMA Intern Med. 2016;176((5)):635–42. doi: 10.1001/jamainternmed.2015.8248. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Stevans JM, Bellon JE, Cohen SM, Zhang Y, James AE, Reynolds B. Comparing Advanced Practice Providers and Physicians as Providers of e-Visits. Telemed e-Health. 2015;21((12)):1019–26. doi: 10.1089/tmj.2014.0248. [DOI] [PubMed] [Google Scholar]
- 9.Moth G, Huibers L, Christensen MB, Vedsted P. Drug prescription by telephone consultation in Danish out-of- hours primary care: A population-based study of frequency and associations with clinical severity and diagnosis. BMC Fam Pract. 2014;15((142)):1–7. doi: 10.1186/1471-2296-15-142. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Stolley P D. L, Becker MH, Lasagna L, McEvilla JD, Sloane LM. The Relationship between Physician Characteristics and Prescribing Appropriateness. Med Care. 1972;10((1)):17–28. doi: 10.1097/00005650-197201000-00003. [DOI] [PubMed] [Google Scholar]
- 11.Davidson W, Molloy W, Somers G, Bedard M. Relation between physician characteristics and prescribing for elderly people in New Brunswick. Can Med Assoc J. 1994;150((6)):917–21. [PMC free article] [PubMed] [Google Scholar]
- 12.Eisenberg JM. Physician Utilization: The State of Research about Physicians’ Practice Patterns. Med Care. [Internet] 1985;23((5)):461–83. Available from: https://www.jstor.org/stable/3764984. [PubMed] [Google Scholar]
- 13.Cms. NPI Records. [Internet]. NPPES NPI Registry. [cited 2019Mar11]. Available from: https://npiregistry.cms.hhs.gov/ [Google Scholar]
- 14.Primary Care. [Internet]. AAFP. American Academy of Family Physicians; 2017 [cited 2019Mar11]. Available from: https://www.aafp.org/about/policies/all/primary-care.html. [Google Scholar]
- 15.McDonald JH. Handbook of Biological Statistics. 3rd ed. Baltimore, MD: Sparky House Publishing. 2014:157–164. [Google Scholar]
- 16.Cameron AC, Trivedi PK. Regression Analysis of Count Data. 2nd ed. Cambridge: Cambridge University Press; 2013. (Econometric Society Monographs) [Google Scholar]
- 17.Weinick RM, Bristol SJ, Desroches CM. Urgent care centers in the U.S.: Findings from a national survey. BMC Heal Serv Researc. 2009;9((79)) doi: 10.1186/1472-6963-9-79. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Pitts BSR, Carrier ER, Rich EC, Kellermann AL. Where Americans Get Acute Care: Increasingly, It’s Not at Their Doctor’ s Office. Health Aff. 2010;29((9)):1620–9. doi: 10.1377/hlthaff.2009.1026. [DOI] [PubMed] [Google Scholar]
- 19.Shih A, Davis K, Schoenbaum SC. Organizing the U.S. Health Care Delivery System for High Performance. 2008 [Google Scholar]
- 20.1. McConnochie KM, Roghmann KJ, Goepp J, Herendeen NE, Wood NE, Ahn DS, et al. Differences in Diagnosis and Treatment Using Telemedicine Versus In-Person Evaluation of Acute Illness. Ambul Pediatr. 2006;6((4)):187–95. doi: 10.1016/j.ambp.2006.03.002. [DOI] [PubMed] [Google Scholar]
- 21.Jaye C, Tilyard M. A qualitative comparative investigation of variation in general practitioners’ prescribing patterns. Br J Gen Pract. 2002;52((478)):381–6. [PMC free article] [PubMed] [Google Scholar]
- 22.Bradley C. Decision making and prescribing patterns: a literature review. Fam Pract. 1991;8:276–287. doi: 10.1093/fampra/8.3.276. [DOI] [PubMed] [Google Scholar]
- 23.Fleming-Dutra KE, Hersh AL, Shapiro DJ, et al. Prevalence of Inappropriate Antibiotic Prescriptions Among US Ambulatory Care Visits, 2010-2011. JAMA: The Journal of the American Medical Association. 2016;315((17)):1864–73. doi: 10.1001/jama.2016.4151. [DOI] [PubMed] [Google Scholar]