The impact of virtual care on antibiotic prescribing practices for Urinary Tract Infections (UTIs): a propensity-score matched cohort study

doi:10.21203/rs.3.rs-8041551/v1

Download PDF

Research Article

The impact of virtual care on antibiotic prescribing practices for Urinary Tract Infections (UTIs): a propensity-score matched cohort study

https://doi.org/10.21203/rs.3.rs-8041551/v1

This work is licensed under a CC BY 4.0 License

You are reading this latest preprint version

Background

Since the emergence of COVID-19, virtual alternatives to in-person care have become commonplace with many physicians providing both virtual and in-person visit options within their practice. However, the rapid implementation of virtual care raises questions regarding its quality compared to in-person care, including for prescribing appropriateness and adherence to clinical guidelines. We examined whether prescribing patterns differed between virtual and in-person physician visits for urinary tract infections (UTIs) in British Columbia, Canada, a clinical presentation for which there was thorough guidance available on first-line therapy.

Methods

We used administrative data from 2022 to examine the association of virtual care with antibiotic prescribing for UTIs, including the likelihood of an antibiotic prescription, the type of antibiotic dispensed (broad- vs. narrow-spectrum), and the duration of antibiotic dispensed. Our analysis used propensity score matching techniques paired with logistic and linear regression models.

Findings:

Compared to in-person visits, virtual visits for UTI were associated with significantly lower odds of antibiotic dispensing (OR = 0.889; 95% CI: 0.872–0.905), significantly higher odds of broad-spectrum antibiotic dispensing (OR = 1.057; 95% CI: 1.015–1.102), and no significant difference in the duration of nitrofurantoin treatment (estimate = 0.337 days; 95% CI: − 0.485 to 1.159).

Interpretation:

Virtual care was associated with slightly lower antibiotic dispensation rates and similar treatment durations, but was also associated with more dispensation of broader spectrum antibiotics prescription. This raises potential concerns regarding antibiotic resistance that merit confirmation in studies that can assess indication, other clinical areas, and other settings.

The COVID-19 pandemic led health care systems worldwide to rapidly expand the use of virtual care, defined here as patient visits with physicians via telephone, video call, or secure messaging.^1,2 In the Canadian province of British Columbia (BC), healthcare providers were instructed to move most consultations to virtual modalities in late March 2020 coinciding with the declaration of COVID-19 as a public health emergency.³ Expanded fee payment codes for virtual care were introduced at the same time to enable this change.⁴

The impact on the use of virtual care was immediate and substantial. In 2018, only 8% of the Canadian healthcare visits were conducted virtually.⁵ By mid-2020, 54% of all routine healthcare visits and 71% of all primary care visits in Canada were conducted virtually.^6,7 Even though the proportion of virtual care visits declined after the COVID-19 pandemic, falling to 30% by March, 2022 as in-person visits resumed,⁶ the share of visits conducted virtually remained substantially higher than pre-pandemic levels.

The rapid implementation and scale-up of virtual care as a stop-gap pandemic measure may have impacted quality of care.⁸ Virtual care has been linked to inappropriate prescribing and clinical prescribing guideline deviations, as limited assessment capabilities may lead physicians to rely on clinical discretion, potentially resulting in unnecessary or prolonged treatments.^9,10 This is a particular concern for antibiotic prescribing where such treatments risk raising population-level antibiotic resistance.^11–13 Broad-spectrum antibiotics (those which are effective against a wide range of bacteria) in particular are often seen as a proxy for increasing resistance in a community, over their more conservative narrow-spectrum (those effective against a more restricted range of bacteria) counterpart.^14–16 First-line treatment often regimens often involve narrow-spectrum antibiotics.¹⁷ In 2021, the Canadian Medical Association (CMA) released the Virtual Care Playbook to help Canadian physicians optimize quality of care in virtual patient encounters and contains a section on the types of conditions that can be safely addressed virtually.¹⁸ Such scope recommendations are intended to help providers curb the need to adjust the types and duration of antibiotics prescribed for the specified infectious conditions, potentially limiting inappropriate antibiotics prescription and antibiotic resistance.

Global evidence on prescribing patterns and adherence to clinical guidelines in virtual care following the COVID-19 pandemic is mixed. Prescription rates and guideline adherence have been highly variable.^19–27 Existing evidence also primarily relies on cross-sectional designs using basic multivariate regression, which may be prone to bias.^19–26 This evidence is additionally primarily based data from the United States data, which may limit generalizability in the Canadian context due to the much more established history of virtual care in the US.

To provide methodologically rigorous evidence in other health systems, including those which have historically relied less on virtual care, the present study aimed to evaluate differences in prescribing rates, spectrum of antibiotics prescribed, and days of medication prescribed for Urinary Tract Infections (UTIs) in British Columbia in 2022. Healthcare visits for suspected UTIs are considered to be amenable to virtual care by the CMA due to predominant reliance on patient history alone for diagnosis.¹⁸ First-line treatment involves narrow-spectrum antibiotics, namely nitrofurantoin, which are specifically recommended for their management due to their effective empirical coverage of bacteria in UTIs.¹⁷ Therefore, we used the UTIs and attendant nitrofurantoin prescription to explore the association between virtual care and antibiotic prescribing behaviours. We used a quasi-experimental approach to further improve methodological rigour.

Data Sources

Population Data BC provides deidentified administrative data capturing individual-level health and health-services related information on nearly all 5.4 million BC residents.²⁸ Access to this data provided by the Data Stewards is subject to approval but can be requested for research projects through the Data Stewards or their designated service providers. This study used six datasets provided by Population Data BC. The Medical Services Plan (MSP) Consolidation File provided demographic information, such as age, location of residence, and administrative sex, for all individuals insured under BC’s publicly funded provincial health insurance plan²⁹ The MSP (Payment Information) file provided data on all medically necessary services provided by physicians and nurse practitioners through the province's fee-for-service system, including corresponding International Classification of Diseases-9 (ICD-9) diagnostic codes and billing codes.³⁰ The Discharge Abstracts Database (DAD) provided data on acute care discharges and day surgery cases and the corresponding ICD-10 diagnostic codes.³¹ PharmaNet provided data on all prescription drug dispensations from BC pharmacies and physicians³² The Vital Events and Statistics Births and Deaths datasets included all births and deaths registered in the province which were utilized to restrict the cohort to only those alive during the full study period.^33,34 All inferences, opinions, and conclusions drawn in this publication are those of the authors, and do not reflect the opinions or policies of the Data Stewards.

Study Period

We focused on physician visits and drug dispensations from January 1, 2022, through December 31, 2022 when the influence of the COVID-19 pandemic and its associated distancing measures had largely diminished.³⁵ By early 2022, parallel virtual and in-person care had become the norm for many physicians. Data from 2020 and 2021 was analyzed to capture baseline medication dispensation patterns, physician visits for UTIs, and comorbidities.

Study Population

Our study population included all BC residents diagnosed with a UTI during the study period. We restricted our analysis to visits conducted by family physicians and differentiated between individuals who received care virtually and those who received care in-person. Individuals covered through federally funded programs, such as First Nations Health Benefits, were excluded as our dataset did not include their prescription drug use.

We defined a UTI diagnosis as an outpatient encounter in the MSP file with a urinary infection-related ICD-9 diagnostic code. Encounters were categorized as virtual or in-person based on BC fee-item codes that indicate mode of service delivery. These diagnostic and fee-item codes can be found within the supplemental material [see supplemental material Tables S1 – S3]. These billing codes distinguish between virtual and in-person visits, but are unable to identify the specific virtual care modalities (e.g., telephone versus video call).

We linked prescription drug dispensations from PharmaNet to a physician visit for UTI if the medications were filled within two weeks following a family physician visit. A two-week window was selected because an estimated 82% of patients in BC fill their prescriptions within two weeks of a primary care encounter.³⁶ Moreover, given the acuteness of UTI symptoms, there it is unlikely that patients would delay filling antibiotics prescriptions beyond this timeframe.^37,38

Outcomes of Interest

We examined changes in the following outcomes per visit: 1) the likelihood of antibiotic dispensation; 2) the likelihood of broad-spectrum antibiotic dispensation; and 3) the number of days of nitrofurantoin treatment dispensed. As antibiotics are not always indicated for treatment of UTI-like symptoms and nitrofurantoin, a narrow-spectrum antibiotic, is recommended for its treatment when antibiotics are indicated, these outcomes allowed us to assess prescribing patterns at increasing levels of granularity. We defined broad- and narrow-spectrum antibiotics utilizing a list of all common antibiotics and their spectrum classification obtained from the British Columbia Centre for Disease Control (BCCDC). Figure 1 shows the analytic cohort size and flow of observations as the data were matched and subset by outcome.

Statistical Analysis

We used propensity score matching (PSM) to construct cohorts of patients that had virtual and in-person primary care visits for UTI. Propensity-score (PS) methods are powerful commonly used tools in observational studies to reduce bias and confounding when estimating treatment effects. The PS represents the probability of receiving the treatment of interest given a set of observed baseline characteristics. By matching individuals with similar PS values, treated and untreated observations can be assembled into comparable groups that are balanced on measured covariates.³⁹ PS is particularly useful in our study as the availability of comprehensive administrative data allowed for accurately match our two groups and mitigate confounding by indication. We constructed three PSM models using “greedy” matching and evaluated covariate balance for each cohort corresponding to our three outcomes. A standardized mean difference of 0.1 was used as the threshold to indicate a negligible difference between the treatment and control groups, following standard guidelines.³⁹ Our final PSM model for all cohorts used greedy

1:1 matching with a caliper of 0.25 of the logit of the propensity score, incorporating the following covariates: sex, year of birth, most responsible physician (whether the physician attending the UTI-related visit was the individual's most commonly seen physician), binary indicators for prior UTI visit and UTI-related dispensation in the two years prior to the visit of interest, BC health service delivery area (HSDA) (16 geographic subdivisions of BC health regions), neighborhood income decile as defined by the MSP Consolidation File, Charlson Comorbidity Index as a continuous measure, and the proportion of visits conducted virtually by the physician attending the UTI-related visit. For the type and days of antibiotic prescribed outcomes, the number of prescriptions dispensed in the two years prior to a dispensation for UTI in 2022 was also included.

Variables were selected based on their established association with virtual care including sex, year of birth, geographic location, and income.^40–42 Additional variables (i.e., Charleson comorbidity index, proportion of visits conducted virtually by the attending physician) were included based on their likely influence on the outcomes, informed by clinical and empirical considerations.

We applied three Generalized Estimating Equation (GEE) regression models on the matched dataset: two GEE logistic models assessing the association between virtual care and (1) antibiotic dispensation and (2) broad vs. narrow spectrum antibiotic use for UTIs, and one GEE linear model examining the number of nitrofurantoin days dispensed. All GEE models assumed an exchangeable correlation structure with observations clustered by patient and practitioner. Outcomes included a binary indicator for any antibiotic dispensation, a binary indicator for antibiotic spectrum (broad vs. narrow), and a numeric variable for duration of nitrofurantoin. Each model included the same covariates used in the PSM to ensure consistency and account for any residual confounding.

The analyses in this study were performed using SAS Version 9.4. This study was approved by The University of British Columbia Behavioural Research Ethics Board (certificate number: H24-00177).

Sensitivity Analyses

The sensitivity analyses used the full unmatched dataset, as well as inverse probability of treatment weighting (IPTW) followed by the same analysis.⁴³ IPTW is an alternative PS method in which the inverse of the PS is used as statistical weights to create a sample that upweights samples with lower probability of occurring as predicted by the covariates, and vice versa. This helps balance confounding across exposure groups, enabling an unbiased estimate of the treatment-outcome relationship.³⁹ These methods allowed us to assess the robustness of our results to the unmatched data as well as different PS-based approaches.

Tables 1, 2, and 3 present the characteristics of the odds of antibiotic dispensation, odds of broad-spectrum antibiotic dispensation, and days of nitrofurantoin dispensed cohorts, respectively. In all cohorts, between 21% − 32% of in-person and 61% − 63% of virtual observations were lost after matching.

Across unmatched cohort samples, females appeared to use more virtual care. Individuals who utilized virtual care had more comorbidities, were more likely to have had a UTI-related dispensation and UTI-related visit with their most responsible physician, and their physicians had a higher proportion of visits conducted virtually as opposed to in-person.

Table 1
Characteristics of BC patients who had an UTI-related physician visit between Jan-Dec 2022.
	Before Matching		After Matching
Characteristic	In-Person, N = 184,720¹	Virtual, N = 368,936¹	In-Person, N = 144,955¹	Virtual. N = 144,955¹
Sex
Female	117,622 (64%)	258,619 (70%)	95,031 (66%)	94,402 (65%)
Male	67,098 (36%)	110,317 (30%)	49,924 (34%)	50,553 (35%)
Age	61 (21)	64 (21)	62 (21)	63 (21)
Weighted Charlson Comorbidity Index	1.09 (1.73)	1.32 (1.93)	1.10 (1.74)	1.15 (1.73)
Most Responsible Physician	37,615 (20%)	96,012 (26%)	31,977 (22%)	33,433 (23%)
Physician Virtual Visit Rate	0.52 (0.21)	0.68 (0.14)	0.60 (0.14)	0.60 (0.15)
Prior UTI Dispensation	77,324 (42%)	162,329 (44%)	60,743 (42%)	62,186 (43%)
Prior UTI Visit	99,242 (54%)	206,670 (56%)	77,379 (53%)	79,403 (55%)
BC Health Services Delivery Era
East Kootenay	4,026 (2.2%)	4,983 (1.4%)	3,540 (2.4%)	3,275 (2.3%)
Kootenay Boundary	3,544 (1.9%)	4,205 (1.1%)	2,606 (1.8%)	2,517 (1.7%)
Okanagan	20,158 (11%)	37,988 (10%)	17,181 (12%)	17,129 (12%)
Thompson Cariboo Shuswap	8,464 (4.6%)	13,237 (3.6%)	6,993 (4.8%)	6,995 (4.9%)
Fraser East	16,632 (9%)	32,282 (8.8%)	12,952 (8.9%)	13,134 (9.1%)
Fraser North	21,914 (12%)	48,829 (13%)	17,830 (12%)	17,849 (12%)
Fraser South	31,706 (17%)	79,675 (22%)	24,750 (17%)	26,074 (18%)
Richmond	4,390 (2.4%)	12,395 (3.4%)	3,414 (2.4%)	3,265 (2.3%)
Vancouver	21,313 (12%)	37,984 (10%)	13,495 (9.3%)	13,807 (9.5%)
North Shore–Coast Garibaldi	12,081 (6.5%)	23,186 (6.3%)	9,802 (6.8%)	9,389 (6.5%)
South Vancouver Island	12,495 (6.8%)	32,487 (8.8%)	10,925 (7.5%)	10,951 (7.6%)
Central Vancouver Island	14,696 (8%)	24,503 (6.6%)	11,477 (7.9%)	11,307 (7.8%)
North Vancouver Island	5,513 (3.0%)	7,993 (2.2%)	4,269 (2.9%)	4,177 (2.9%)
Northwest	2,206 (1.2%)	3,583 (1.0%)	1,846 (1.3%)	1,662 (1.1%)
Northern Interior	3,530 (1.9%)	3,970 (1.1%)	2,388 (1.6%)	2,319 (1.6%)
Northeast	1,832 (1.0%)	1,565 (0.4%)	939 (0.6%)	931 (0.6%)
Unknown Region	221 (0.1%)	278 (0.1%)	132 (< 0.1%)	110 (< 0.1%)
Neighbourhood Income Decile
1	18,167 (9.9%)	25,112 (6.8%)	11,620 (8.1%)	12,504 (8.7%)
2	18,769 (10%)	37,650 (10%)	15,023 (10%)	15,407 (11%)
3	21,129 (12%)	37,325 (10%)	16,711 (12%)	16,465 (11%)
4	21,429 (12%)	42,907 (12%)	17,240 (12%)	17,161 (12%)
5	22,718 (12%)	47,994 (13%)	19,063 (13%)	19,063 (13%)
6	17,799 (9.6%)	40,205 (11%)	14,043 (9.7%)	13,877 (9.6%)
7	19,514 (11%)	42,878 (12%)	15,640 (11%)	15,619 (11%)
8	17,019 (9.3%)	34,646 (9.4%)	13,710 (9.5%)	13,576 (9.4%)
9	15,587 (8.4%)	33,941 (9.2%)	12,932 (8.9%)	12,873 (8.9%)
10	10,792 (5.9%)	23,926 (6.5%)	8,876 (6.1%)	8,744 (6.1%)
Unknown	932 (0.5%)	767 (0.2%)	525 (0.4%)	552 (0.4%)
¹n (%), Mean (SD)
Note: Due to rounding, percentages may not add to 100%

Table 2
Characteristics of BC patients who had an UTI-related physician visit between Jan-Dec 2022 and were dispensed an antibiotic virtually or in-person
	Before Matching		After Matching
Characteristic	In-Person, N = 43,289¹	Virtual, N = 76,929¹	In-Person, N = 30,248¹	Virtual. N = 30,248¹
Sex
Female	35,381 (82%)	63,336 (84%)	24,872 (82%)	24,855 (82%)
Male	7,908 (18%)	12,593 (16%)	5,376 (18%)	5,393 (18%)
Age	54 (22)	56 (22)	56 (22)	57 (22)
Weighted Charlson Comorbidity Index	0.68 (1.38)	0.80 (1.50)	0.72 (1.40)	0.72 (1.37)
Most Responsible Physician	6,692 (15%)	15,730 (20%)	5,333 (18%)	5,627 (19%)
Physician Virtual Visit Rate	0.49 (0.21)	0.69 (0.15)	0.59 (0.14)	0.59 (0.14)
Prior UTI Dispensation	13,225 (31%)	27,170 (35%)	9,885 (33%)	10,030 (33%)
Prior UTI Visit	16,401 (38%)	32,336 (42%)	12,045 (40%)	12,143 (40%)
Dispensations in Previous Two Years	96 (425)	106 (486)	98 (417)	101 (412)
BC Health Services Delivery Era
East Kootenay	1,126 (2.6%)	909 (1.2%)	740 (2.4%)	724 (2.4%)
Kootenay Boundary	1,068 (2.5%)	992 (1.3%)	649 (2.1%)	652 (2.2%)
Okanagan	4,928 (11%)	7,244 (9.4%)	3,527 (12%)	3,536 (12%)
Thompson Cariboo Shuswap	2,039 (4.7%)	3,018 (3.9%)	1,632 (5.4%)	1,622 (5.4%)
Fraser East	3,900 (9.0%)	6,571 (8.5%)	2,813 (9.3%)	2,859 (9.5%)
Fraser North	5,060 (12%)	10,927 (14%)	3,764 (12%)	3,831 (13%)
Fraser South	6,262 (14%)	15,089 (20%)	4,682 (15%)	4,692 (16%)
Richmond	867 (2.0%)	2,142 (2.8%)	602 (2.0%)	566 (1.9%)
Vancouver	4,541 (10%)	8,482 (12%)	2,696 (8.9%)	2,700 (8.9%)
North Shore–Coast Garibaldi	3,324 (7.7%)	5,026 (6.5%)	1,575 (5.2%)	1,542 (5.1%)
South Vancouver Island	1,843 (4.3%)	6,463 (8.4%)	2,688 (8.9%)	2,650 (8.8%)
Central Vancouver Island	4,253 (9.8%)	5,411 (7.0%)	1,193 (7.9%)	1,191 (3.9%)
North Vancouver Island	1,823 (4.2%)	1,889 (2.5%)	506 (1.7%)	483 (1.6%)
Northwest	654 (1.5%)	1,063 (1.4%)	506 (1.7%)	483 (1.6%)
Northern Interior	936 (2.2%)	1,046 (1.4%)	587 (1.9%)	602 (2.0%)
Northeast	632 (1.5%)	595 (0.8%)	297 (1.0%)	307 (1.0%)
Unknown Region	33 (< 0.1%)	72 (< 0.1%)	18 (< 0.1%)	18 (< 0.1%)
Neighbourhood Income Decile
1	2,644 (6.1%)	3,840 (5.0%)	1,745 (5.8%)	1,739 (5.8%)
2	3,790 (8.8%)	6,002 (7.8%)	2,656 (8.8%)	2,651 (8.8%)
3	4,451 (10%)	7,253 (9.5%)	3,008 (10%)	3,008 (10%)
4	5,091 (12%)	8,604 (12%)	3,518 (12%)	3,451 (11%)
5	5,534 (13%)	9,685 (13%)	3,851 (13%)	3,914 (13%)
6	4,644 (11%)	8,768 (11%)	3,227 (11%)	3,210 (11%)
7	5,073 (12%)	9,406 (12%)	3,583 (12%)	3,563 (11%)
8	4,514 (10%)	8,781 (11%)	3,246 (11%)	3,283 (11%)
9	4,247 (9.9%)	8,321 (11%)	3,079 (10%)	3,093 (10%)
10	2,961 (6.9%)	5,840 (7.6%)	2,145 (7.1%)	2,141 (7.1%)
Unknown	127 (0.3%)	132 (0.2%)	63 (0.2%)	68 (0.2%)
¹n (%), Mean (SD)
Note: Due to rounding, percentages may not add to 100%

Table 3
Characteristics of BC patients who had an UTI-related physician visit between Jan-Dec 2022 and were dispensed nitrofurantoin virtually or in-person
	Before Matching		After Matching
Characteristic	In-Person, N = 18,301¹	Virtual, N = 33,406¹	In-Person, N = 12,408¹	Virtual. N = 12,408¹
Sex
Female	17,136 (94%)	31,098 (93%)	11,568 (93%)	11,533 (93%)
Male	1,165 (6.4%)	2,308 (6.9%)	840 (6.8%)	875 (7.1%)
Age	52 (21)	53 (21)	54 (21)	54 (21)
Weighted Charlson Comorbidity Index	0.56 (1.29)	0.64 (1.33)	0.59 (1.29)	0.61 (1.25)
Most Responsible Physician	2,644 (14%)	6,206 (19%)	2,036 (16%)	2,155 (17%)
Physician Virtual Visit Rate	0.48 (0.22)	0.70 (0.15)	0.59 (0.14)	0.59 (0.14)
Prior UTI Dispensation	4,278 (23%)	9,444 (28%)	3,167 (26%)	3,224 (26%)
Prior UTI Visit	5,849 (32%)	12,035 (36%)	4,202 (34%)	4,259 (34%)
Dispensations in Previous Two Years	80 (367)	79 (327)	78 (349)	83 (310)
BC Health Services Delivery Era
East Kootenay	452 (2.5%)	392 (1.2%)	331 (2.7%)	300 (2.4%)
Kootenay Boundary	406 (2.2%)	433 (1.3%)	260 (2.1%)	256 (2.1%)
Okanagan	2,101 (11%)	2,720 (8.1%)	1,343 (11%)	1,359 (11%)
Thompson Cariboo Shuswap	864 (4.7%)	1,348 (4.0%)	665 (5.4%)	693 (5.6%)
Fraser East	1,335 (7.3%)	2,812 (8.4%)	1,016 (8.2%)	1,062 (8.6%)
Fraser North	2,222 (12%)	4,975 (15%)	1,619 (13%)	1,636 (13%)
Fraser South	2,751 (15%)	6,791 (20%)	2,018 (16%)	2,028 (16%)
Richmond	362 (2.0%)	960 (2.9%)	243 (2.0%)	233 (1.9%)
Vancouver	2,304 (13%)	4,010 (12%)	1,241 (10%)	1,217 (9.8%)
North Shore–Coast Garibaldi	1,421 (7.8%)	2,046 (6.1%)	906 (7.3%)	915 (7.4%)
South Vancouver Island	654 (3.6%)	2,798 (8.4%)	569 (4.6%)	524 (4.2%)
Central Vancouver Island	1,682 (9.2%)	2,194 (6.6%)	1,093 (8.8%)	1,102 (8.9%)
North Vancouver Island	746 (4.1%)	672 (2.0%)	469 (3.8%)	448 (3.6%)
Northwest	281 (1.5%)	486 (1.5%)	220 (1.8%)	217 (1.7%)
Northern Interior	423 (2.3%)	431 (1.3%)	281 (2.3%)	275 (2.2%)
Northeast	282 (1.5%)	305 (0.9%)	127 (1.0%)	135 (1.1%)
Unknown Region	15 (< 0.1%)	33 (< 0.1%)	7 (< 0.1%)	8 (< 0.1%)
Neighbourhood Income Decile
1	1,083 (5.9%)	1,605 (4.8%)	702 (5.7%)	705 (5.7%)
2	1,495 (8.2%)	2,326 (7.0%)	990 (8.0%)	960 (7.8%)
3	1,853 (10%)	3,171 (9.5%)	1,227 (9.9%)	1,240 (10%)
4	2,095 (12%)	3,613 (11%)	1,369 (11%)	1,369 (11%)
5	2,302 (13%)	4,307 (13%)	1,582 (13%)	1,618 (13%)
6	2,015 (11%)	3,925 (12%)	1,359 (11%)	1,350 (11%)
7	2,186 (12%)	4,157 (12%)	1,520 (12%)	1,506 (12%)
8	1,991 (11%)	3,898 (12%)	1,396 (11%)	1,413 (11%)
9	1,837 (10%)	3,687 (11%)	1,284 (10%)	1,281 (10%)
10	1,305 (7.2%)	2,534 (7.6%)	906 (7.3%)	896 (7.3%)
Unknown	46 (0.3%)	61 (0.2%)	22 (0.2%)	19 (0.2%)
¹n (%), Mean (SD)
Note: Due to rounding, percentages may not add to 100%

After matching, no notable differences were seen between in-person and virtual visits across cohorts. Figures of the distributions of the PS in the matched and unmatched cohorts and the inverse propensity scores in the IPTW cohorts can be found within the supplemental material [see supplemental material Figures S1 – S9].

The regression analysis on the matched cohort showed that virtual care visits for UTI resulted in significantly lower odds of antibiotic dispensing (OR = 0.889; 95% CI: 0.872–0.905; p = < 0.0001). For the average individual in our matched cohorts, this translated to a predicted probability of receiving an antibiotic prescription of 34.4% (95% CI: 33.3–35.4%) for virtual visits and 35.4% (95% CI: 29.2–42.2%) for in-person visits.

Despite the overall lower odds of prescribing, virtual care visits resulted in significantly higher odds of broad-spectrum antibiotic dispensing (OR = 1.057; 95% CI: 1.015–1.102; p = 0.0081). This corresponded to a predicted probability of receiving a broad-spectrum antibiotic of 16.1% (95% CI: 14.7–17.5%) for virtual visits and 15.5% (95% CI: 8.7–26.1%) for in-person visits.

Finally, we found no significant difference in the duration of nitrofurantoin treatment (estimate = 0.337 days; 95% CI: − 0.485 to 1.159; p = 0.4121) when compared to in-person visits. These model results translated to an average predicted duration of nitrofurantoin dispensed of 6.18 days (95% CI: 5.25–7.12) for virtual visits and 7.79 days (95% CI: 3.53–12.04) for in-person visits for the average cohort member.

Sensitivity Analysis

The IPTW analysis produced results largely consistent with our primary analysis. Compared to in-person visits, virtual care visits resulted in lower odds of antibiotic dispensing (OR = 0.849; 95% CI 0.828 to 0.871; p < 0.0001), higher odds of broad-spectrum antibiotic dispensing (OR = 1.067; 95% CI 1.024 to 1.113; p = 0.0022), and no significant difference in the duration of nitrofurantoin treatment (estimate = -0.798; 95% CI -1.651 to 0.055; p = 0.0667).

Sensitivity analysis utilizing the full unmatched cohort found that, compared to in-person visits, virtual care visits for UTI resulted in significantly lower odds of antibiotic dispensing (OR = 0.932; 95% CI 0.918 to 0.947; p < 0.0001), significantly greater odds of broad-spectrum antibiotic dispensing (OR = 1.064; 95% CI 1.028 to 1.103; p = 0.0005), and no significant difference in the duration of nitrofurantoin treatment (estimate = -0.048; 95% CI -0.568 to 0.473; p = 0.8573).

Virtual care was associated with a small but statistically significant decrease in the likelihood of antibiotic dispensation, a small but statistically significant increase in of broad-spectrum antibiotic use, and no significant difference in the duration of nitrofurantoin treatment compared to in-person care. These findings suggest that virtual care may not necessarily lead to greater overall antibiotic use or longer treatment durations; however, when antibiotics are prescribed through virtual care, physicians may prefer broader spectrum options.

Despite these differences, the average individual with a virtual care visit for UTI in our dataset had a predicted 34.43% probability of receiving an antibiotic, just 1.01% lower than for the same individual with an in-person visit. The same individual had only a 0.59% increased probability of being dispensed a broad-spectrum antibiotic.

Our sensitivity analysis with IPTW yielded values of slightly different magnitude and significance from the primary analysis in the likelihood of antibiotic and broad-spectrum antibiotic dispensation outcomes. These differences may be attributed, in part, to the large loss of virtual care observations in both PSM cohorts, leading to a different sample. However, the full unmatched cohort produced a result similar to the PSM cohort for both outcomes, suggesting that the IPTW estimates may have been biased by extreme inverse propensity score weights as seen in Figures S3, S6, and S9.

Given that we found that antibiotics were less likely to be dispensed for a UTI in virtual verses in-person visits, and treatment duration did not differ between the two service modes, our results suggest that virtual care does not contribute to antibiotic overprescribing at a greater level than in-person care. These findings may help to alleviate concerns that limited patient evaluation in virtual settings leads to increased risk of antibiotic resistance through inappropriate antibiotic prescribing. However, our finding of a potentially higher likelihood of broad-spectrum antibiotic dispensation suggests that virtual care may be associated with reduced adherence to clinical guidelines, therefore warranting additional assessment with larger samples and in other settings.

Our results of the likelihood of dispensing an antibiotic and difference in the duration of nitrofurantoin treatment for UTI support two distinct bodies of evidence: one linking virtual care to reduced antibiotic prescribing,^20,21 and another showing no change in prescribing patterns.^22–25 Our findings on the type of antibiotics dispensed, which suggest potentially lower guideline concordance in virtual care, contrast with existing evidence indicating no difference in^21,22 or greater adherence to^21,22 guideline-concordant prescribing practices during virtual visits compared to in-person visits. However, our results derived from what we believe is a more robust quasi-experimental methodological approach than utilized in the existing literature are likely to offer more reliable evidence. Nonetheless, even with this stronger design, the predicted probabilities of all outcomes differed only slightly between virtual and in-person visits, bringing into question the practical impact of the observed changes in dispensation patterns for UTI.

Limitations

As this study utilized population-level administrative health records, we were limited to assessing the drugs dispensed rather than those prescribed. As such, we were unable to directly observe the clinician’s prescribing decision. Dispensed records may differ from prescriptions written, as pharmacists can adapt, substitute, or modify prescriptions based on clinical judgment, formulary restrictions, or patient-specific factors. We assumed based on previous research that no antibiotic dispensed within two weeks of a UTI visit implied no prescription. However, this may have led to misclassification if prescriptions were dispensed outside that timeframe or were unrelated to the visit, particularly among patients with multiple comorbidities. This likely had the greatest impact on the likelihood of antibiotic dispensation outcome. Additionally, our quantitative modeling, used a variety of administrative databases that rely on the term “sex”, with the response options “female”, “male” or “unknown”; whether this is reflects sex assigned at birth, legal sex, or gender cannot be determined

Limitations also stem from our propensity score matching. While we matched on many variables using a narrow caliper to ensure comparability, unmeasured confounding may still exist, and the exclusion of unmatched individuals could explain differences between the matched and IPTW models. Additionally, clinical factors like UTI history beyond the study period were unavailable and may have influenced treatment decisions. Finally, subsetting the cohort for the second and third outcomes reduced the sample size, particularly for nitrofurantoin users, which would have reduced our statistical power for detecting differences in antibiotic type and duration.

We found that, for UTI-related visits in British Columbia, virtual care does not appear to increase the risk of antibiotic resistance with respect to overall antibiotic dispensation rates or the duration of nitrofurantoin treatment, as these were slightly lower or comparable to in-person visits. However, our finding that virtual care was associated with a greater likelihood of broad-spectrum antibiotic dispensation may indicate a deviation from UTI clinical prescribing guidelines and potential increased risk of antibiotic resistance in virtual care. If confirmed in other studies and settings, these results suggest that current prescribing practices in virtual care settings are largely appropriate, however, intervention may be required in terms of the type of antibiotic prescribed. Should these findings be robust, they indicate virtual care is being used generally in line with clinical guidelines, but further assessment is needed in clinical areas where guidelines are less well defined.

Human Ethics and Consent to Participate

This study was approved by The University of British Columbia Behavioural Research Ethics Board (certificate number: H24-00177). This study used fully anonymized administrative data. Individual consent to participate was not required

Consent for Publication

Not Applicable

Authors’ Contributors

MR, LH, KM, ST, AS and MRL conceptually designed the study. MR, LH, KM, LC, and MRL designed the methodology. MR, AS, and LC curated and cleaned the data, performed the statistical analysis, and validated and visualized the results. MR drafted the manuscript. All authors critically edited, read, and approved the final manuscript. MR and LC had full access to all the data in the study and had responsibility for the integrity of the data and the accuracy of the data analysis.

Availability of data and materials

The code for the data cleaning analysis model is available at a GitHub repository (hosted at https://github.com/maryannrogers/Virtual-Care). Access to the Population Data BC datasets for individual level administrative health data can be obtained through submitting a request at https://www.popdata.bc.ca/forms/intake_meeting.

Competing Interests

Dr. Law has consulted for Health Canada and Canada’s Drug Agency and acted as an expert witness for the Durham Police Association, the Society of United Professionals, and Orr Taylor LLP. The other authors declare no competing interests.

Funding

This study received funding from the Canadian Institutes for Health Research (F22-01982). Michael Law received salary support from a Canada Research Chair in Access to Medicines. Lindsay Hedden and Sian Tsuei are respectively funded by a Scholar Award and Research Award from Michael Smith Health Research British Columbia.

Acknowledgements

The authors would like to thank the BCCDC Community Antimicrobial Stewardship for provision of their list of commonly used antibiotics as well as Sandra Peterson for her analytic advice.

Rawaf S, Allen,Luke N, Stigler, Florian L et al (2020) Kringos, Dionne, Quezada Yamamoto, Harumi, van Weel, Chris,. Lessons on the COVID-19 pandemic, for and by primary care professionals worldwide. European Journal of General Practice. ;26(1):129–33
Matenge S, Sturgiss E, Desborough J, Hall Dykgraaf S, Dut G, Kidd M (2022) Ensuring the continuation of routine primary care during the COVID-19 pandemic: a review of the international literature. Fam Pract 39(4):747–761
Henry B COVID-19: Important Update from the Provincial Health Officer [Internet]. 2020 [cited 2024 Aug 5]. Available from: https://www.doctorsofbc.ca/sites/default/files/covid-19_important_update_from_the_provincial_health_officer.pdf
Doctors of BC (2020) Fee Guide Revisions & Updates. Oct
Canadian Medical Association Virtual Care in Canada: Discussion Paper [Internet]. 2018 Dec [cited 2024 Aug 5]. Available from: https://www.cma.ca/sites/default/files/pdf/News/Virtual_Care_discussionpaper_v2EN.pdf
Canada Health Infoway. Canadians’ Health Care Experiences During COVID-19: Uptake of Virtual Care (2022) ; Available from: https://www.infoway-inforoute.ca/en/component/edocman/3828-canadians-health-care-experiences-during-covid-19/view-document?Itemid=103
Glazier RH, Green ME, Wu FC, Frymire E, Kopp A, Kiran T (2021) Shifts in office and virtual primary care during the early COVID-19 pandemic in Ontario, Canada. Can Med Assoc J 193(6):E200
Canada H Virtual care policy framework [Internet]. 2022 [cited 2024 Sept 3]. Available from: https://www.canada.ca/en/health-canada/corporate/transparency/health-agreements/bilateral-agreement-pan-canadian-virtual-care-priorities-covid-19/policy-framework.html
Hui D, Dolcine B, Loshak H Approaches to Evaluations of Virtual Care in Primary Care [Internet]. Canada’s Drug Agency; 2022 Jan [cited 2025 Feb 16]. Available from: https://www.cda-amc.ca/evaluations-virtual-care
Hardcastle L, Ogbogu U Virtual care: Enhancing access or harming care? Healthc Manage Forum 2020 July 20;33(6):288–292
Pfister T, Schröder S, Heck J, Bleich S, Krüger THC, Wedegärtner F et al Potentially inappropriate prescriptions of antibiotics in geriatric psychiatry—a retrospective cohort study. Frontiers in Psychiatry [Internet]. 2024 [cited 2024 Mar 3];14. Available from: https://www.frontiersin.org/journals/psychiatry/articles/10.3389/fpsyt.2023.1272695
MacGowan A, Macnaughton E (2017) Antibiotic resistance. Medicine 45(10):622–628
Chokshi A, Sifri Z, Cennimo D, Horng H (2019) Global Contributors to Antibiotic Resistance. J Global Infect Dis 11(1):36
Melander J, Zurawski RV, Melander D (2018) Narrow-spectrum antibacterial agents. MedChemComm 9(1):12–21
Cižman M, Plankar Srovin T (2018) Antibiotic consumption and resistance of gram-negative pathogens (collateral damage). GMS Infect Dis 6:Doc05
Niazi-Ali S, Bircher J (2022) Broad spectrum antibiotic stewardship by quality improvement methods. Int J Risk Saf Med 33(S1):S35–40
Kashouris E, Joseph A, Lewis T (2023) Nitrofurantoin: what is the evidence for current UK guidance? J Antimicrob Chemother 78(11):2605–2611
Canadian Medical Association, College of Family Physicians of Canada, Royal College of Physicians and Surgeons of Canada. Virtual Care Playbook [Internet]. Ottawa: The Association (2021) Available from: https://digitallibrary.cma.ca/link/digitallibrary52
Penza KS, Murray MA, Myers JF, Maxson J, Furst JW, Pecina JL (2020) Treating pediatric conjunctivitis without an exam: An evaluation of outcomes and antibiotic usage. J Telemed Telecare 26(1–2):73–78
Miller LE, Bhattacharyya N (2021) Antibiotic Prescribing for Acute Rhinosinusitis: In-Person Versus Virtual Visits During Covid-19. Laryngoscope 131(7):E2121–E2124
Ray KN, Martin JM, Wolfson D, Schweiberger K, Schoemer P, Cepullio C et al (2021) Antibiotic Prescribing for Acute Respiratory Tract Infections During Telemedicine Visits Within a Pediatric Primary Care Network. Academic Pediatrics. Sept 1;21(7):1239–43
Johnson KL, Dumkow LE, Salvati LA, Johnson KM, Yee MA, Egwuatu NE (2021) Comparison of diagnosis and prescribing practices between virtual visits and office visits for adults diagnosed with uncomplicated urinary tract infections within a primary care network. Infect Control Hosp Epidemiol 42(5):586–591
Li C, Ong C, Morris A, Woollons I, Ashfaq A, Jagatia R (2021) Evaluating the Appropriateness of Antibiotic Treatment of Tonsillitis during COVID-19 in the North Wale Primary Healthcare Setting. J Prim Care Community Health 12:21501327211003687
Murray MA, Penza KS, Myers JF, Furst JW, Pecina JL (2020) Comparison of eVisit Management of Urinary Symptoms and Urinary Tract Infections with Standard Care. Telemed J E Health 26(5):639–644
Bakhit M, Baillie E, Krzyzaniak N, van Driel M, Clark J, Glasziou P et al (2021) Antibiotic prescribing for acute infections in synchronous telehealth consultations: a systematic review and meta-analysis. BJGP Open. ;5(6):BJGPO.2021.0106
Cuellar A, Pomeroy JML, Burla S, Jena AB (2022 Sept) Quality of Antibiotic Prescribing in a Large Direct-to-Patient Telehealth Program: an Observational Study. J GEN INTERN MED 1(12):3202–3204
Ostberg N, Ip W, Brown I, Li R (2022) Impact of telemedicine on clinical practice patterns for patients with chest pain in the emergency department. Int J Med Informatics 161:104726
Data linkage | Population Data BC [Internet]. [cited 2025 Mar 28]. Available from: https://www.popdata.bc.ca/data_linkage
Central Demographics File (MSP Registration and Premium Billings Client Roster and Census Geodata)/Consolidation file (MSP registration and premium billing) data set | Population Data BC [Internet]. [cited 2025 Mar 28]. Available from: https://www.popdata.bc.ca/data/demographic/consolidation_file
Medical Services Plan data set | Population Data BC [Internet]. [cited 2025 Mar 28]. Available from: https://www.popdata.bc.ca/data/health/msp
Discharge Abstract Database (Hospital Separations) data set | Population Data BC [Internet]. [cited 2025 Mar 28]. Available from: https://www.popdata.bc.ca/data/health/dad
PharmaNet data set | Population Data BC [Internet]. [cited 2025 Mar 28]. Available from: https://www.popdata.bc.ca/data/health/pharmanet
BC Vital Events and Statistics Deaths data set | Population Data BC [Internet]. [cited 2025 Mar 28]. Available from: https://www.popdata.bc.ca/data/demographic/vs_deaths
BC Vital Events and Statistics Births data set | Population Data BC [Internet]. [cited 2025 Mar 28]. Available from: https://www.popdata.bc.ca/data/demographic/vs_births
COVID-19 Update for February 15 2022 [Internet]. 2022 [cited 2025 June 9]. Available from: https://www.youtube.com/watch?v=oZdwMrfhyFI
Zeitouny S, Cheng L, Wong ST, Tadrous M, McGrail K, Law MR (2023) Prevalence and predictors of primary nonadherence to medications prescribed in primary care. Can Med Assoc J 195(30):E1000–E1009
Acute Cystitis Causes, Symptoms & Treatment [Internet]. Cleveland Clinic. [cited 2025 May 1]. Available from: https://my.clevelandclinic.org/health/diseases/24450-acute-cystitis
Overview Acute cystitis. In: InformedHealth.org [Internet] [Internet]. Institute for Quality and Efficiency in Health Care (IQWiG); 2023 [cited 2025 May 1]. Available from: https://www.ncbi.nlm.nih.gov/books/NBK279403/
Haukoos JS, Lewis RJ (2019) The Propensity Score. In: Livingston EH, Lewis RJ, editors. JAMA Guide to Statistics and Methods [Internet]. New York, NY: McGraw-Hill Education; Available from: jamaevidence.mhmedical.com/content.aspx?aid = 1184195125
Rush KL, Seaton CL, Corman K, Hawe N, Li EPH, Dow-Fleisner SJ et al (2022) Virtual Care Prior to and During COVID-19: Cross-sectional Survey of Rural and Urban Adults. JMIR Formative Res 6(8):e37059
Liu L, Goodarzi Z, Jones A, Posno R, Straus SE, Watt JA (2021) Factors associated with virtual care access in older adults: a cross-sectional study. Age and Ageing. July 1;50(4):1412–5
Mistry SK, Shaw M, Raffan F, Johnson G, Perren K, Shoko S et al (2022) Inequity in Access and Delivery of Virtual Care Interventions: A Scoping Review. Int J Environ Res Public Health 19(15):9411
Scott L, Thiebaud P (2007) Using Propensity Scores to Adjust For Treatment Selection Bias. In: Statistics and Data Analysis [Internet]. Available from: https://support.sas.com/resources/papers/proceedings/proceedings/forum2007/184-2007.pdf

Competing interest reported. Dr. Law has consulted for Health Canada and Canada’s Drug Agency and acted as an expert witness for the Durham Police Association, the Society of United Professionals, and Orr Taylor LLP. The other authors declare no competing interests.

SupplementalMaterial.docx

Download PDF

Reviews received at journal
13 Dec, 2025
Reviews received at journal
12 Dec, 2025
Reviewers agreed at journal
04 Dec, 2025
Reviewers agreed at journal
04 Dec, 2025
Reviewers invited by journal
04 Dec, 2025
Editor invited by journal
11 Nov, 2025
Editor assigned by journal
10 Nov, 2025
Submission checks completed at journal
10 Nov, 2025
First submitted to journal
05 Nov, 2025

You are reading this latest preprint version

The impact of virtual care on antibiotic prescribing practices for Urinary Tract Infections (UTIs): a propensity-score matched cohort study

Status:

Version 1

Abstract

Background

Methods

Findings:

Interpretation:

Figures

Background

Methods

Statistical Analysis

Results

Discussion

Conclusion

Declarations

References

Additional Declarations

Supplementary Files

Status:

Version 1