Modelling the risk of SARS-CoV-2 infection through PPE doffing in a hospital environment

Self-contamination during doffing of personal protective equipment (PPE) is a concern for healthcare workers (HCW) following SARS-CoV-2 positive patient care. Staff may subconsciously become contaminated through improper glove removal, so quantifying this risk is critical for safe working procedures. HCW surface contact sequences on a respiratory ward were modelled using a discrete-time Markov chin for: IV-drip care, blood pressure monitoring and doctors' rounds. Accretion of viral RNA on gloves during care was modelled using a stochastic recurrence relation. The HCW then doffed PPE and contaminated themselves in a fraction of cases based on increasing case load. The risk of infection from this exposure was quantified using a dose-response methodology. A parametric study was conducted to analyse the effect of: 1a) increasing patient numbers on the ward, 1b) the proportion of COVID-19 cases, 2) the length of a shift and 3) the probability of touching contaminated PPE. The driving factors for infection risk were surface contamination and number of surface contacts. HCWs on a 100% COVID-19 ward were less than 2-fold more at risk than on a 50% COVID ward (1.6% vs 1%), whilst on a 5% COVID-19 ward, the risk dropped to 0.1% per shift (sd=0.6%). IV-drip care resulted in higher risk than blood pressure monitoring (1.1% vs 1% p<0.0001), whilst doctors' rounds produced a 0.6% risk (sd=0.8%). Recommendations include supervised PPE doffing procedures such as the "doffing buddy" scheme, maximising hand hygiene compliance post-doffing and targeted surface cleaning for surfaces away from the patient vicinity.


58
Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) is an enveloped 59 virus which has infected in excess of 10 million people to date and caused more 60 than 700,000 deaths worldwide according to Johns Hopkins University's COVID-19 61 Dashboard (2). Inanimate objects known as fomites may host pathogens and have 62 the potential to contribute to infection transmission in healthcare environments. This 63 occurs in viral infection spread (3)(4)(5) including 7). There appears to be 64 similarity between persistence of SARS-CoV-1 and 2 on surfaces, with viable virus 65 shown to be present for up to 72 hours (8). This allows an opportunity for exposure 66 through hand-to-fomite contacts, especially if surfaces are heavily contaminated. 67 Although personal protective equipment (PPE) such as gloves, gowns, and masks 68 are worn to protect both patient and healthcare worker (HCW) from exposure, self-69 contamination during PPE doffing processes (9, 10) poses risks to HCW and enables 70 spread from one patient to another during multiple care episodes. SARS-CoV-2 has 71 been detected on healthcare worker PPE (11) and in the environment of rooms 72 where doffing occurs, providing evidence that errors in doffing could facilitate 73 COVID-19 exposure and transmission. 74 While SARS-CoV-2 has been detected on PPE and patient surfaces, the 75 relationship between viral RNA concentrations and risk of infection is still 76 unknown(12). Quantitative microbial risk assessments (QMRA) involve the use of 77 mathematical models to estimate doses of a pathogen and subsequent infection 78 risk probabilities. Quantifying infection risk for any given dose can be used to guide 79 intervention decision-making and have been used in other public health contexts, 80 such as in setting water quality standards (13). These typically rely on experimental 81 doses of a microorganism inoculated into healthy participants or mice models in a 82 known quantity. Whether they develop the infection can then be recorded(13). 83 . CC-BY 4.0 International license It is made available under a perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted September 23, 2020. . https://doi.org/10.1101/2020 that the number of care episodes per shift would be less than for high load 110 conditions. The assumed number of patient care episodes when PPE is worn per shift 111 for low and high case load scenarios were 7 and 14, respectively, based on a 112 respiratory ward in a university teaching hospital in the UK. The low case load 113 estimate was based on communication with a UK NHS consultant, who tracked the 114 number of gowns used by healthcare workers over a week on a mixed  bed respiratory ward. All model parameters are described and reported in Table 1 Hand hygiene efficacy: alcohol gel (log10 reduction) Uniform (min=2, max=4) Hand hygiene efficacy: soap and water (log10 reduction) Normal (mean=1.62, sd=0.12) Left-and right-truncated at 0 and 6, respectively . CC-BY 4.0 International license It is made available under a perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted September 23, 2020. . https://doi.org/10.1101/2020  is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted September 23, 2020. . https://doi.org/10.1101/2020 random manner, insofar that moving from one surface category to another has a 141 higher probability than a transition elsewhere. By assigning each surface category a 142 numerical value from 1 to 5, where Equipment = 1, Patient = 2, Hygiene 143 areas = 3, Near-bed surfaces = 4, and Far-bed surfaces = 5, HCW sequential contact 144 of surfaces can be modelled in terms of weighted probabilities (14). 145 The movement of a HCW between surfaces is modelled using a discrete-time 146 Markov chain approach (14). Using defined weighted probabilities based on 147 observation of patient care, surface contact by HCW can be simulated based on 148 the property that, given the present state, the future and past surfaces touched are 149 independent. This is termed the Markov property (eq 1): 150 (1) 151 Where represents the surface contacted in the ℎ event, and are two 152 surfaces, and represents a conditional probability. This is then denoted → for 153 ease of notation. For example, the probability if the HCW is currently touching the 154 table that they will next touch the chair is Discrete-time Markov chains were fitted to observed care contact sequences 159 using the "markovchainFit" function from the R package markovchain (version 160 0.7.0). Separate Markov chains were fitted to IV care, doctors' rounds and 161 observational care sequences. States included "in" (entrance to the patient room), 162 "out" (exit from the patient room), contact with a far-patient surface, contact with a 163 near-patient surface, contact with a hygiene surface (e.g. tap, sink, soap or alcohol 164 dispenser), and contact with equipment. For each episode of care, the first event 165 was entrance into the patient room. All HCWs wore a gown, gloves, mask and face 166 . CC-BY 4.0 International license It is made available under a perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted September 23, 2020.  (14) to respond to a recurrence relationship which the concentration 172 on hands after the n th contact, ℎ , with the concentration on hands, −1 ℎ , and on 173 the surface involved, −1 , before the contact. See eq. 2. 174 This is an adaptation of the pathogen accretion model (PAM) from King et al. 176 (2015) (14) and a gradient transfer model by Julian et al. (2009) (30). Here, the 177 concentration on hands for contact n is equal to the previous concentration on the 178 hand ( −1 ℎ ) after adjusting for inactivation for the virus on the hand ( ℎ ) and surface 179 , minus the removal from the hand due to hand-to-surface transfer plus the gain to 180 the hand due to surface-to-hand transfer. Δ is the time taken for an episode of 181 patient care and sampled from a uniform distribution of range 2-20minutes(31). 182 Here, and represent hand-to-surface and surface-to-hand transfer efficiencies 183 respectively. The fraction of the total hand surface area ( ℎ ) is used to estimate how 184 much virus is available for transfer given a concentration of number of viral 185 particles/cm 2 on the gloved hand and surface. 186 is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint

Estimating Inactivation on the Hand
The copyright holder for this this version posted September 23, 2020. . https://doi.org/10.1101/2020.09.20.20197368 doi: medRxiv preprint therefore used a uniform distribution with a minimum of 1 hour and a maximum of 8 193 hours to estimate a distribution of ℎ inactivation rates. 194

195
The decay of the virus causing COVID-19 has been shown to vary under both 196 humidity and temperature but in contrast with previous findings(8), it appears that 197 surface material may not have a significant impact on decay rate(33). We therefore 198 take a conservative approach and use an averaged half-life estimate for stainless 199 steel and plastic-coated surfaces at 21-23C(8) at 40% relative humidity; which are 200 5.63h (95%CI=4.59-6.86h), and 6.81h (95%CI=5.62-8.17h), respectively. We assume a 201 first order decay (eq 3) to estimate inactivation constant which we use here for is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted September 23, 2020.

234
If the patient was assumed to be infected, surface contamination levels (RNA/ 235 swab surface area) were sampled from a triangular distribution where the minimum 236 and maximum were informed by minimum and maximum contamination levels 237 reported for surfaces in an intensive care unit ward (1). The median of these was 238 used to inform the midpoint of the triangular distribution (1). For patient contacts, the 239 concentration of virus detected on a patient mask was used as a point value (3.3 x 240 10 3 RNA/swab surface area) (1). When a patient was not infected, it was assumed 241 . CC-BY 4.0 International license It is made available under a perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted September 23, 2020. . https://doi.org/10. 1101/2020 contacts with surfaces and with the patient would not contribute to additional 242 accretion of virus on gloved hands. 243 Surface areas to relate RNA/swabbed surface area to RNA/cm 2 were not 244 provided by Guo et al. (2020). While a typical sampling size is 100 cm 2 , it may be as 245 small as 10-25 cm 2 (35-38) and in real-world scenarios, sampling surface areas may 246 be larger or smaller than these depending upon available surface area, ease of 247 access and the contamination magnitude expected. Since the surface areas of 248 these surfaces were not provided, a triangular distribution (min=5, max=195, 249 mid=100) describing the surface area (cm 2 ) of surfaces sampled was used to 250 estimate RNA/cm 2 . Not all detected RNA was assumed to represent infectious viral 251 particles. This is a conservative risk approach when utilizing molecular concentration 252 data in QMRA (39). Therefore, concentrations on surfaces (viable viral 253 particles/cm 2 ) were estimated by eq 4, 254 where is the RNA/swabbed surface area, is the surface area (cm 2 ) of 256 the surface, and is the fraction of RNA that relates to infective viral particles 257 (uniform(min=0.001, max=0.1)). 258

259
For all scenarios, it was assumed the starting concentration on gloved hands for 260 the first episode of care was equal to 0 viral particles/cm 2 . If gloves were doffed and 261 a new pair was donned in between care episodes, it was assumed the next episode 262 of care began with a concentration of 0 viral particles/cm 2 on the gloved hands. 263 After each care episode, a number was randomly sampled from a uniform 264 distribution with a minimum of 0 and a maximum of 1. If this value was less than or 265 equal to the set probability of self-contamination during doffing, self-contamination 266 . CC-BY 4.0 International license It is made available under a perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted September 23, 2020. . https://doi.org/10.1101/2020.09.20.20197368 doi: medRxiv preprint occurred, where the fraction of total virus transferred from the outer glove surface to 267 the hands was assumed to be uniformly distributed between 3 x 10 -5 % and 10% (9). 268 There was then a 50/50 chance that either hands were washed or sanitized using 269 alcohol gel. If they washed their hands, a log10 reduction was randomly sampled 270 from a normal distribution with a mean of 1.62 and a standard deviation of 0.12, 271 (min=0 and max=6) (25). While these are not coronavirus-specific hand washing 272 efficacies they allow for a conservative estimate. If hand sanitizer was used, a log10 273 reduction was randomly sampled from a uniform distribution with a minimum of 2 274 and a maximum of 4 (24). 275 To estimate a dose, an expected concentration on the hands after doffing and 276 hand hygiene was estimated, followed by an expected transfer to a facial mucosal 277 membrane during a single hand-to-nose contact after each patient care episode 278 (eq. 5). 279 There was a 50/50 chance that either the right or left hand was used for this 281 hand-to-face contact. Here, the transfer efficiency (TH  M) of the hand-to-nose 282 contact was randomly sampled from a normal distribution with a mean of 33.90%, 283 and a standard deviation of 20% based on a viral surrogate (21) is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted September 23, 2020. . https://doi.org/10.1101/2020.09.20.20197368 doi: medRxiv preprint simulation was used.
represents the time between doffing and touching the 293 mucosa. 10,000 parameter combinations are obtained for each care type scenario 294 in a Monte Carlo framework. 295

296
Due to lack of dose-response curve data for SARS-CoV-2, an exact beta-297 Poisson dose-response curve (40) was fitted to pooled data for SARS-CoV-1 and 298 HCoV 229E, assuming the infectivity of SARS-CoV-2 lies between the infectivity for 299 these two organisms. In eq 6., 1F1( , + , − ) is the "Kummer confluent 300 hypergeometric function" and ( ) is the probability of infection risk given dose: 301 (eq. 6) (40). 302 303 Ten-thousand bootstrapped pairs of and were produced based on a maximum 304 likelihood estimation fit. For each estimated dose, an and pair were randomly 305 sampled, and an infection risk was estimated with eq. 6. The infectious dose for 50% 306 of infections to occur was between 5 and 100 infectious viral particles with a mean 307 of 30; the dose-response curve can be seen in Figure 1. 308 . CC-BY 4.0 International license It is made available under a perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted September 23, 2020. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted September 23, 2020. At a 7-patient load, regardless of COVID-19 prevalence, risk was 0.6% whilst doubling 332 patient capacity more than doubled the risk to 1.3% per shift. Figure 3 shows a bar 333 chart with standard deviations for care type, COVID-19 prevalence on the ward 334 and chance of self-contamination following a mistake during doffing. 335 . CC-BY 4.0 International license It is made available under a perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted September 23, 2020.

339
Risks relating to COVID prevalence on the ward does not appear to track 340 linearly. HCWs on a 100% COVID ward were less than a 2x more at risk than on a 50% 341 COVID ward (1.6% vs 1%), whilst on a 5% COVID, the risk dropped to 0.1% per shift on 342 average (sd=0.6%). 343 In terms of most important factor determining risk, Figure 4 shows two dimensional 344 heatmaps of input parameters plotted against predicted infection risk to elucidate 345 correlations. The stronger the correlation, the more influence that parameter has on 346 the output. Surface cleanliness was found to be the single most important factor in 347 determining future risk, with hand-to-mouth/eyes/nose transfer efficiency only half as 348 important (correlation coefficient = 0.29 vs = 0.12, respectively) (see Table 2). 349 Surface concentration relates to cleaning frequency and hence the control case 350 . CC-BY 4.0 International license It is made available under a perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted September 23, 2020. . https://doi.org/10.1101/2020 was run for half the surface bioburden. At double the cleaning frequency, the risk is 351 halved. 352 is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted September 23, 2020. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted September 23, 2020. . https://doi.org/10. 1101/2020 high-touch surfaces(46) that had historically been prioritised for cleaning, may need 390 to be revised. 391 Regardless of the number of COVID-19 positive patients on a ward, notable 392 decreases in predicted infection risk were associated with less self-contamination 393 during doffing. For example, for scenarios involving all COVID-19 patients, the mean 394 infection risk for 10% probability of self-contamination while doffing was 0.4%, while 395 the mean infection risk for an 80% probability of self-contamination while doffing was 396 more than a 420% increase at 2.1% (Table 1). This emphasizes the importance of 397 adequate training for PPE use. Less risk of self-contamination will decrease 398 transmission risks, potentially through a doffing buddy. PPE can be an effective 399 strategy for mitigating exposure if proper doffing techniques are used. In addition to 400 training, improvements in PPE design that enhance safety and expediency of 401 doffing may lower self-contamination rates and therefore improve PPE as a 402 mitigation strategy (47). For example, fasteners or ties on gowns/masks were 403 identified as "doffing barriers," because it was unclear whether these were to be 404 untied and there were difficulties in reaching these ties. Self-contamination due to 405 gowns and masks were not specifically addressed in this model. It is possible that 406 self-contamination during doffing of items other than gloves could increase 407 potential risks due to incorrect doffing. Shortages of PPE have changed normal 408 practice where PPE is worn on a sessional basis rather than renewed for each 409 patient. This means less doffing and potentially less auto-contamination but may 410 increase the risk of virus transfer within the unit. 411 In addition to the importance of safe and proper doffing, the results from this 412 computational study also emphasize the importance of surface decontamination 413 and environmental monitoring strategies. The concentration of virus on surfaces was 414 the most influential parameter on infection risk, which is consistent with other surface 415 . CC-BY 4.0 International license It is made available under a perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted September 23, 2020. . https://doi.org/10. 1101/2020 transmission risk studies (30). Whilst SARS-CoV-2 RNA has been detected on surfaces, 416 one limitation to a molecular approach is the lack of information regarding 417 infectivity. In a recent study by Zhou et al. (2020), no surface samples demonstrated 418 infectivity. However, it was noted that the concentrations of SARS-CoV-2 on surfaces 419 were below the current detection limits for culture methodologies ( According to CDC, of 428,295 healthcare personnel for which data were available, 433 20% (84,035/428,295) were COVID-19 cases (51). However, it is not known how many 434 shifts were associated with these infection rates. Additionally, we assumed that 435 wards with non-COVID-19 patients did not have SARS-CoV-2 contamination on 436 surfaces, due to lack of data on SARS-CoV-2 surface contamination beyond COVID-437 19 wards or patient rooms. There is potential for asymptomatically infected 438 healthcare workers to contribute environmental contamination, especially when 439 considering the relatively long shedding durations for asymptomatic infections (52). 440 . CC-BY 4.0 International license It is made available under a perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted September 23, 2020. . https://doi.org/10.1101/2020 Infected healthcare workers and environmental contamination could be 441 considered in future extensions of this model. 442 The fact that the proportions of healthcare workers with COVID-19 discussed 443 above are much larger than the infection risks estimated suggest that other 444 transmission routes could drive additional HCW cases. This would include more 445 transmission through airborne routes, or HCW to HCW transmission by asymptomatic 446 cases outside the COVID-19 care environment (53). However, while there continues 447 to be disagreement over the contribution of each route to overall risk, transmission 448 routes influence each other, making them all significant in healthcare environments. 449 For example, surfaces can become contaminated due to deposition of aerosolized 450 virus. Viruses can later be resuspended from surfaces, contributing to air 451 contamination. Future work should extend current models with a multi-exposure 452 pathway approach. This will advance not only our understanding of SARS-CoV-2 453 transmission but the transmission of pathogens in built environments as a whole. 454 Finally, a dose-response curve informed by SARS-CoV-1 and HCoV 229E data 455 was utilized, due to lack of SARS-CoV-2-specific dose-response data. We suggest 456 that this therefore is a conservative estimate. Despite limitations related to dose-457 response, the conclusions from the estimated doses were consistent with insights 458 from infection risk estimates. Increases in probability of contamination between care 459 episodes related to increases in dose and most notably, for scenarios in which more 460 than 5% of patients had COVID-19 (Figure 3, Figure S1). 461 is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted September 23, 2020. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted September 23, 2020. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted September 23, 2020. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted September 23, 2020. . https://doi.org/10.1101/2020.09.20.20197368 doi: medRxiv preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted September 23, 2020. . https://doi.org/10. 1101/2020