Waiting times, patient flow, and occupancy density in South African primary health care clinics: implications for infection prevention and control

Background Transmission of respiratory pathogens, such as Mycobacterium tuberculosis and severe acute respiratory syndrome coronavirus 2, is more likely during close, prolonged contact and when sharing a poorly ventilated space. In clinics in KwaZulu-Natal (KZN) and Western Cape (WC) provinces, South Africa, we estimated clinic visit duration, time spent indoors and outdoors, and occupancy density of waiting rooms. Methods We used unique barcodes to track attendees' movements in 11 clinics in two provinces, multiple imputation to estimate missing arrival and departure times, and mixed-effects linear regression to examine associations with visit duration. Results 2,903 attendees were included. Median visit duration was 2 hours 36 minutes (interquartile range [IQR] 01:36-3:43). Longer mean visit times were associated with being female (13.5 minutes longer; p<0.001) and attending with a baby (18.8 minutes longer; p<0.01), and shorter mean times with later arrival (14.9 minutes shorter per hour after 0700; p<0.001) and attendance for tuberculosis or ante/postnatal care (24.8 and 32.6 minutes shorter, respectively, than HIV/acute care; p<0.01). Overall, attendees spent more of their time indoors (median 95.6% [IQR 46-100]) than outdoors (2.5% [IQR 0-35]). Attendees at clinics with outdoor waiting areas spent a greater proportion (median 13.7% [IQR 1-75]) of their time outdoors. In two clinics in KZN (no appointment system), occupancy densities of ~2.0 persons/m2 were observed in smaller waiting rooms during busy periods. In one clinic in WC (appointment system), occupancy density did not exceed 1.0 persons/m2 despite higher overall attendance. Conclusions Longer waiting times were associated with early arrival, being female, and attending with a young child. Attendees generally waited where they were asked to. Regular estimation of occupancy density (as patient flow proxy) may help staff assess for risk of infection transmission and guide intervention to reduce time spent in risky spaces.

practice in most clinics, though the methods used vary by province. National guidelines recommend that 57 total visit times should be less than three hours. [29] 58 Most approaches to estimating waiting times conceptualise the patient's journey through the clinic as linear, 59 with each individual passing through the clinic as quickly as possible while ensuring that the necessary 'touch 60 points' are accessed. In South Africa, waiting times are usually measured through the provision of a physical 61 all individuals attending together (denoted by cards having been stapled together). Data from questionnaires 128 administered to facility managers were entered into a password-protected Excel spreadsheet. 129 Data were excluded from this analysis where clinic entry and exit were not recorded (because the research 143 team was too small to monitor all entrances and exits of the clinic). Using multiply-imputed data (n = 20 144 imputations), relationships between individual characteristics and time spent in clinic (continuous outcome) 145

Analysis
were examined using a mixed-effects linear regression model with a random effect for clinic. Province was 146 included as fixed effect. The shape of the relationship between time of arrival and time spent in clinic was 147 examined using fractional polynomials regression with a set of defined powers (−2, −1, −0.5, 0.5, 1, 2, and 148 ln [x]) and a maximum of two power terms in the model. The differences in model deviances were compared: 149 the linear model was used if the improvement in fit was not statistically significant at p <0.05. Province, age 150 group, sex, and the ratio of patients to clinical staff were included in the multivariable model as a priori 151 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 23, 2021. ; https://doi.org/10.1101/2021.07.21.21260806 doi: medRxiv preprint confounders; other variables were included if they showed an important association (p <0.05) in the 152 univariable model. Coefficients, representing the difference in mean time spent in clinic (in minutes), are 153 reported with 95% confidence intervals (CIs). 154

Proportion of time spent indoors vs. outdoors
155 Non-imputed data were used from clinics where a scanner had been positioned at all facility entrance/s and 156 all indoor/outdoor doorways. Individuals with a total captured visit time of less than five minutes were 157 excluded, as they were considered likely to have discarded their barcode. Each individual's pathway through 158 the clinic was mapped: for each barcode scan recorded, the individual's location in the time preceding the 159 scan was categorised as 'indoors', 'outdoors', or 'unknown' (if they appeared to have moved between two 160 unconnected locations, indicating a missing barcode scan) based on the location of their previous scan. Total 161 time spent in each type of location (as a proportion of the individual's overall recorded time in clinic) was 162 summarised by clinic and by self-reported reason for clinic attendance. 163

Occupancy density
164 Non-imputed data were used from clinics that had more than one indoor waiting area and where a barcode 165 scanner had been positioned at all entrances and exits of at least two waiting areas. Data were divided into 166 10 second slices and entries and exits from each demarcated space noted for each 10 second period; the 167 number of individuals within a space at the end of each 10 second period was divided by the floor area and 168 volume of that space to give the occupancy density (in persons/m 2 and persons/m 3 , respectively) for that 10 169

180
Patient flow in study clinics was broadly organised around three key steps in the following order: 1) patient 181 registration (file collection); 2) vital signs; and 3) HCW consultation. Individuals usually waited in different 182 parts of the clinics for each step. The pathway taken depended on the reason for visit (many individuals also 183 visited one or more of the in-clinic pharmacy, phlebotomist, and other specialist practitioners) and was 184 implemented variably in clinics based on their size, design, and organisation of care. In most clinics, 185 individuals attending for TB care (i.e., those being investigated for TB or taking anti-TB treatment) were 'fast-186 tracked' and skipped steps 1 and 2 above. Clinics varied widely in size, population served, services offered, 187 and organisation of care. Importantly, some clinics routinely asked patients to wait in covered outdoor 188 waiting areas, whereas others had only indoor areas. All clinics in WC and no clinics in KZN operated a date-189 time appointment system for at least some patients (Supplementary table 5 is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted July 23, 2021. ; https://doi.org/10.1101/2021.07.21.21260806 doi: medRxiv preprint collected, this could not be rectified, and the two categories were combined in analysis (but are shown 201 separately in Table 1). 202 Table 1. Characteristics of clinics, individuals attending, and staff working on the day of data collection, overall  In univariable analysis (Table 2), there was strong evidence of an increase in mean time spent in clinic for 212 individuals who were female (p <0.001), attending with a baby (p <0.001), or attending with ≥1 other person 213 (p <0.01). There was also strong evidence of differences by reason for visit (p <0.01): individuals attending 214 for TB care and ante/post-natal care spent the shortest time in clinic. Mean time in clinic reduced by ~15 215 minutes for each hour that arrival was delayed after 0700 (p <0.001). 216 Table 2. Results of univariable and multivariable mixed-effects linear regression using imputed data, showing 217 effects of different factors on total time spent in clinic (n = 2,634; 11 exercises in 10 clinics) 218 In multivariable analysis, longer mean times remained associated with being female (13.5 [95% CI 6-21] 219 minutes longer than males) and attending with a baby (18.8 [95% CI 8-30] minutes longer than those 220 attending without). Reason for visit (p <0.01) and time of arrival (p <0.001) also remained important: those 221 attending for TB care or ante/post-natal care spent a mean 24.8 (95% CI 9-41) minutes and 32.6 (95% CI 11-222 54) minutes less in clinic, respectively, than those attending for HIV/acute care, and mean time in clinic 223 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 23, 2021. ; https://doi.org/10.1101/2021.07.21.21260806 doi: medRxiv preprint reduced by 14.9 (95% CI 13-17) minutes for each hour that arrival was delayed after 0700. The results of the 224 fractional polynomial models showed that the linear model adequately described the relationship between 225 the time at clinic and arrival time (Appendix 3.2.1). 226 Occupancy density of indoor spaces 242 Data from three clinics were sufficient to estimate occupancy density of at least three indoor spaces ( Figure  243 2). In clinic KZN6 (Figure 2, panel 2), the occupancy density of area A consistently declined over the course of 244 the day as individuals moved into areas B and C. Because of its relatively large volume, the occupancy 245 density of area A never went above 0.9 persons/m 2 . In contrast, in the smallest space (area C), occupancy 246 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 23, 2021. ; https://doi.org/10.1101/2021.07.21.21260806 doi: medRxiv preprint peaked at around 1200, with density around or above 2.0 persons/m 2 from 1000-1200. In clinic KZN2 (panel 247 3), the smaller overall numbers of attendees meant that although the spaces are of similar size to clinic 248 KZN6, density was generally lower. Overall occupancy was highest in clinic WC1, but the larger waiting 249 spaces in this clinic meant that occupancy density was never higher than 1.0 persons/m 2 (panel 4), even in 250 the smallest space. Clinic WC1 also had a well-functioning date-time appointment system, which is likely why 251 occupancy of these spaces was more evenly distributed over the day compared with the other two clinics. 252 Occupancy density by room volume (persons/m 3 ) was calculated for the same spaces (Supplementary table  256   9). This is a more relevant measure of occupancy density for predominantly airborne pathogens, such as 257 Mtb. All assessed waiting spaces in clinics KZN2 and KZN6 had relatively low ceilings (maximum height 2.5-258 2.7 m) and occupancy density was higher (median 0.21-1.02 persons/m 3 ) than in spaces in clinic WC1, where 259 ceilings were higher (maximum height 4.2-5.9 m; median occupancy density 0.10-0.14 persons/m 3 ). 260

261
We tracked 2,903 clinic attendees at 11 PHC clinics in two provinces of South Africa. Median time spent in 262 clinic was 2 hours 36 minutes (IQR 01:36-03:43). People who arrived early in the morning spent longer in 263 clinic, as did women and individuals attending with babies. Individuals attending for TB and maternal care 264 spent less time in clinic. People attending clinics that had outdoor covered waiting areas spent more of their 265 visit time outdoors, though differences were also seen between individuals attending the same clinic based 266 on how care was organised for different 'streams'. In clinics with multiple indoor waiting areas, occupancy 267 was often not distributed evenly between areas or over time; periods of high occupancy density (>2 268 persons/m 2 ) were observed in smaller waiting areas. 269 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 23, 2021. 'preventive' and 'curative' care, respectively). Patterns in our data were also observed by previous 278 investigators, including longer times for individuals who arrived earlier [23,24,42] and the early arrival of the 279 majority of attendees, often before the clinic opened. [30] A higher patient to nurse ratio was strongly 280 associated with longer waiting times in the study by Egbujie et al., [42] but not in our study, possibly because 281 our estimates of staff numbers included all clinical staff, not only nurses. We are not aware of any previous 282 studies that estimated proportions of time spent indoors vs. outdoors or the occupancy density of waiting 283 areas. 284 Early arrival and queueing outside clinics is common in South Africa. It is influenced by the frequent absence 285 of appointment and queue management systems; the organisation of services around the 'morning rush'; 286 the lack of incentives for staff to change working patterns; and complex factors outside the health system, 287 such as the availability of public transport and the community's trust in the system. Detailed exploration of 288 these issues is beyond the scope of this paper, but some discussion can be found in the report of an Umoya 289 omuhle workshop on patient flow that involved a range of South African experts. [43]  sites, [17,45] and reducing the frequency of routine clinic visits for certain conditions, for example by increasing 313 the amount of medication provided (trials among people taking ART have shown promising results). [16,[46][47][48][49] 314 Measures to improve the overall efficiency of the clinic aim to move people through the facility as quickly as 315 possible and to reduce the likelihood of bottlenecks in flow. These include holistic approaches, such as 316 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 23, 2021. ; https://doi.org/10.1101/2021.07.21.21260806 doi: medRxiv preprint 'Lean', [50,51] value-stream mapping, [52] and other quality improvement methods, [53] as well as more targeted 317 changes in staffing or resources at specific points in clinical pathways. [24] 318 Streaming and triage interventions focus on the movement of people once they enter a health facility. In line 319 with Ideal Clinic guidance, [28] every clinic in our study operated a streaming system that allowed people 320 attending for TB care to bypass many of the steps in the pathway. This is partly intended to reduce the risk of 321 Mtb transmission and is made feasible by the relatively small numbers of people treated for TB at each clinic 322 and because no additional triage process is required. Triage (broadly defined as the process of prioritising 323 patients for care based on their needs) [54] has also been shown to reduce waiting times in a hospital in South 324 Africa, though it was less effective when used in two PHC clinics. [55,56] Effective triage can be challenging and 325 resource-intensive to sustain, [57] and sub-optimal implementation of symptom-based triage for TB IPC has 326 been documented by several studies. [58][59][60][61] Active queue management has also been tested: a qualitative 327 study around the use of a 'Fast Queue' in clinics in KZN found that the use of multiple, managed queues was 328 Ethiopia [63] and Kenya, [64] antenatal clinics in Mozambique, [65] and PHC clinics in South Africa, [42] the last as 333 part of a suite of interventions that included streaming, training, and infrastructure upgrades. Investigators 334 describe generally encouraging results, though they also highlight the considerable challenges involved in 335 standardising implementation at facilities that are differently organised. During the Umoya omuhle patient 336 flow workshop, discussions around appointment system implementation emphasised the importance of 337 support processes (such as pre-retrieval of files) and technological infrastructure in sustaining this complex 338 intervention. [43] 339 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

340
Building flexibility into the organisation of flow would allow a clinic to adapt to and absorb periods of 341 increased traffic without putting patients or staff at risk; for example, by moving people from an 342 overcrowded area to one that is relatively empty, or by activating 'overflow' covered outdoor waiting areas. 343 However, any such initiative would require 1) a queue management system, to ensure that individuals 344 moved between areas are not placed at a disadvantage, and 2) clinic managers to have a) easy access to 345 real-time information about flow and b) the resources and freedom to try to improve flow. [27] Patient flow 346 can be difficult to measure quickly: previous published descriptions focus on largely qualitative descriptions 347 of observed movement patterns. [23,66] Occupancy density, however, is easy to measure (e.g., through manual 348 headcounts) and, measured periodically across a clinic, could be used as a proxy estimate for flow. We 349 suggest that regular, light-touch ('diagnostic') approximation of this metric may have numerous potential 350 direct and indirect benefits, including improved efficiency; shorter waiting times; better clinic-specific 351 decision-making; and a strengthened relationship between the clinic and its community. [27,43,67] 352 Importantly, interventions intended to reduce attendance and waiting times may adversely affect the flow 353 around the which the clinic was designed and may therefore increase the rate of transmission to an 354 individual during the time they do spend in the clinic. Most clinics are designed with waiting areas that get 355 successively smaller as patients move through the pathway; as pathways diverge, patients 'diffuse' through 356 the clinic and one would expect occupancy to be lower. However, if the overall 'patient load' is greater than 357 the capacity of the clinic, or if different stages of the pathway are variably efficient, or if certain attendees 358 (e.g., those with appointments) are allowed to skip parts of the queue, bottlenecks can arise in areas that 359 are designed to hold fewer people, leading to higher than optimal occupancy of 'downstream' areas and/or 360 under-use of 'upstream' areas. Interventions to improve flow and reduce waiting times are acutely 361 vulnerable to achieving "many small successes and one big failure" [68] and should be undertaken with careful 362 consideration of potential effects on other parts of the pathway, possible increases in risk of disease 363 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 23, 2021. ; https://doi.org/10.1101/2021.07.21.21260806 doi: medRxiv preprint transmission, and adjustments that may be needed in resource allocation, ventilation, and the organisation 364 of care. 365

366
The method employed in this study was relatively inexpensive, built on methods already widely used in 367 South African PHCs, and included elements that could be incorporated into routine estimation of waiting 368 times and flow. Numbers of individuals who declined to participate were not recorded and we were 369 therefore unable to assess for selection bias introduced by the enrolment process. Starting data collection 370 after some individuals had arrived and stopping data collection at 1400 (because of logistical restrictions) 371 reduced the numbers of individuals whose data could be used to estimate total waiting time, requiring the 372 use of multiple imputation to deal with missing data. Multiple imputation assumes that the data are missing 373 at random, which means that the observed values can be used to predict the missing values. However, if the 374 assumption is incorrect, the results may be biased. Furthermore, the validity of results derived from multiply 375 imputed data depend on the appropriateness of the imputation model. Future exercises should, at a 376 minimum, continue to record clinic exits for as long as possible. Because of variability between and within 377 clinics, and because data were collected on only one day from almost all clinics, estimates presented here 378 should not be considered representative of the two provinces, types of clinics, or the clinics themselves. In 379 busy clinics in particular, many attendees' barcodes were not scanned every time at every scanning point, 380 and estimates of waiting area occupancy and time spent indoors or outdoors should be treated as 381 approximations. Even so, our headline findings are plausible and consistent with those from other studies. 382

383
Measuring patient flow is important for estimating clinic efficiency and disease transmission risk. In our 384 study, women, individuals arriving early, and those attending with young children spent longer at clinic. 385 Attendees generally waited where they were asked to: using outdoor waiting areas as part of patient 386 pathways increased the proportion of visit time spent outdoors. Occupancy of indoor spaces varied 387 considerably over the day and people often were not distributed evenly throughout the available space. 388 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 23, 2021. . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 23, 2021.

550
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 23, 2021.

Figure 1. Box and whiskers plots showing proportions of time spent indoors and outdoors, by clinic (panel 625
A) and for two visits to clinic KZN1, by selected reasons for visit (panel B) 626 627 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted July 23, 2021. ; https://doi.org/10.1101/2021.07.21.21260806 doi: medRxiv preprint Page 29 of 31 *Clinic has at least one outdoor waiting area that is part of the patient pathway. 628 †Clinic has at least one outdoor waiting area, but it is not part of the patient pathway.

629
The central horizontal line represents the median value; boxes represent the interquartile range (IQR); and whiskers 630 represent largest and smallest values within 1.5 IQR of the upper and lower quartiles, respectively. Time spent in 631 unknown locations was negligible for most clinics and is therefore not shown. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted July 23, 2021. ; https://doi.org/10.1101/2021.07.21.21260806 doi: medRxiv preprint Figure 2. Line graph (panel 1) and heat maps (panels 2-4) showing, respectively, total numbers of people 638 in three indoor waiting areas and approximate occupancy density (in persons/m 2 ) of each waiting area in 639 each of clinics KZN2, KZN6, and WC1 from 0800-1345* 640 641 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted July 23, 2021.

644
Total numbers (line graph) indicative only of numbers of people occupying the three spaces examined, not overall 645 numbers of people in the entire clinic.

646
Spaces A, D, and G were the main (pre-filing +/-pre-vitals) formal waiting areas for their respective clinics; spaces B, C, 647 H, and I were formal (pre-vitals and/or pre-consultation) waiting areas; space E was a corridor used as a pre-648 consultation waiting area; and space F was a combined pre-vitals waiting area, vitals administration area, and patient 649 registration area. 650 hh: hours; IQR: interquartile range; mm: minutes 651 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted July 23, 2021. ; https://doi.org/10.1101/2021.07.21.21260806 doi: medRxiv preprint