Estimation of air change rate by CO₂ sensor network in workplace with COVID-19 outbreak

Abstract

Objectives: This study aimed to measure the air change per hour (ACH) in a workplace that spanned 880 m² and had a ceiling height of 3 m. This workplace experienced clusters of coronavirus disease (COVID-19) cases, and the study measured ACH before and after remediation. The objective was to provide a quantitative estimate of ACH in various compartments. Methods: A network of CO₂ sensors was set up in the workplace. The data from the sensors were analyzed using a generalized linear mixed model and dynamic time distortion to estimate the ACH in each area. Results: During the cluster outbreak, the ACH was in the range of 0.408 to 1.178/hour (p<.001), which was relatively low and likely contributed to the outbreak. Additionally, the room’s ventilation was imbalanced due to partitioning. However, the ACH improved significantly from 1.835 to 2.551/hour (p<.001) by simply opening the windows and allowing natural ventilation. Conclusions: Based on the evidence that the transmission of COVID-19 was contained following the enhancement of ventilation, an ACH rate of below 2/hour was the primary factor in developing COVID-19 clusters within the facility under investigation.

Introduction

Controlling the spread of novel coronavirus disease 2019 (COVID-19) has become a priority worldwide. After case reports in December 2019, social distancing has been widely adopted as a containment strategy. Adopting this social lifestyle has reaffirmed the importance of direct human connections and face-to-face interactions¹⁾. Therefore, controlling the risks and ensuring safety in educational centers, the public, and workplaces is essential. This requires the concerted efforts of managers, supervisors, administrators, and all other stakeholders.

On March 28, 2022, the National Institute of Infectious Diseases of Japan announced for the first time that aerosols are a primary route of infection, along with contact and droplet transmission, which had been earlier recognized²⁾. The institute presented the results of an analysis of four actual COVID-19 clusters of presumed aerosol infections, three of which were indoors, in which infectious aerosols were transmitted to remote persons in yoga classes³⁾, shopping malls⁴⁾, and choir choruses⁵⁾, resulting in mass infection. The remaining case of infection in an airplane in Japan⁶⁾ was an aerosol infection cluster, as secondary infectious agents were widely distributed in a narrow-body (3-3 rows) airplane. From the beginning, it has been pointed out that such COVID-19 clusters occur in enclosed spaces with poor ventilation. For example, there are known cases of cluster infections in an extremely poorly ventilated restaurant in China⁷⁾, in the business class of a wide-body (3-3-3 row) airplane⁸⁾, and in a study pointing out the possibility that aerosol infection is dominant on the Diamond Princess Cruise ship⁹⁾. All of these suggest the importance of aerosol infection control measures to prevent the spread of COVID-19.

Measures against all three routes of infection (contact, droplet, and aerosol) can reduce COVID-19 transmission risk through multiple defenses like social distancing, mask-wearing, and vaccination¹⁰⁾. However, compared to contact and droplet transmission, which can be prevented by social distancing and the use of masks, aerosol transmission is difficult to visually recognize, and the effectiveness of respective countermeasures has not been confirmed. Accordingly, mass transmissions of COVID-19 have been reported in poorly ventilated areas¹¹⁾. Additionally, the inappropriate use of plastic sheeting for preventing droplet infection has caused clusters of infectious diseases and threatened workplace safety¹²⁾.

To avoid such risks, using CO₂ sensors to control indoor air quality has attracted significant attention^{13,14,15,16,17)}. Indoor CO₂ concentration (referring to exhaled air) measurement is considered an effective method for indirect risk management to ensure that exhaled aerosol particles containing severe acute respiratory syndrome coronavirus-2 (SARS-CoV-2) do not remain indoors. Therefore, these devices have been widely installed as a safety measure in places where people gather, such as restaurants, stores, classrooms, and offices. The guideline for its operation considers a provisional control value of 800–1,000 ppm, set by government agencies of various countries as the maximum CO₂ concentration¹⁸⁾. Under pandemic conditions, the Centers for Disease Control and Prevention has indicated that the CO₂ concentration should be maintained below 800 ppm¹⁹⁾, which has been considered a good safety indicator.

As CO₂ concentration is not a direct risk indicator and there is no direct epidemiological relationship between CO₂ concentration and COVID-19 transmission, this method’s effectiveness should be verified to define appropriate control measures to reduce transmission risk and improve workplace safety. In the manufacturing industry, quality assurance focuses on processes to ensure quality, and the proper control of a process is an indicator of product quality²⁰⁾. Similarly, COVID-19 risk management requires managing and controlling the environment (ie, ventilation) rather than the CO₂ concentration itself.

Building room ventilation can generally be calculated by dividing the building volume by the ventilation volume of the installed ventilation measures. However, years after building construction, the layout of rooms might change, or the ventilation system’s performance can degrade due to malfunction or lack of maintenance²¹⁾, so the ventilation system’s capacity will not be maintained as calculated. To overcome this issue, the ventilation in a target room can be evaluated and quantitatively assessed based on the CO₂ concentration behavior monitored by CO₂ sensors. Additionally, the ventilation performance is not uniform for floors with large areas and complex layouts^12,22). Therefore, a systematic evaluation should be conducted by analyzing the time series of data from sensors arranged in a network, and these results should be used to improve ventilation and reduce transmission risks.

Herein, we used a CO₂ sensor network and conducted a tracer gas experiment to evaluate the complex ventilation conditions in a business site in Japan, where a cluster of COVID-19 infections occurred. In 2021, 14 infections occurred among 29 crew members who spent a few hours a day in a work preparation area of 880 m² and a ceiling height of 3 m, sharing the same void with 23 desk staff members. The room was divided into four partitions, and only one doorway served as a natural ventilation route. The distance between the crew members was >2 m. The risk of contact and droplet infection was small, while the possibility of aerosol infection was high. However, when the health department conducted an on-site inspection, it was noted that contact infection via business tablets was the cause, so all tablets were disinfected. According to preliminary interviews with the facility manager, the chance of contact infection via tablets was very low, as only one tablet was used by each person and not shared.

Thus, this study focused on aerosol infection and evaluated the air change rate for each partition considering the same conditions under which the COVID-19 cluster occurred and improved conditions in which ventilation routes were determined. Based on this analysis, we performed a quantitative risk assessment. The data from the sensor network were statistically analyzed using generalized linear mixed models (GLMMs) and dynamic time warping to verify the effect of ventilation improvement. Based on the results and our previous investigations, we discussed the ventilation index that reduces the occurrence of COVID-19 clusters. The results suggest that the air change per hour (ACH) improvement is reasonable and consistent with the lack of subsequent reports of infections.

The study aimed to: 1) demonstrate a method for evaluating and determining the state of air quality management in an office with complex geometry using a CO₂ sensor network and 2) verify the effectiveness of ventilation improvement measures in the building where the outbreaks occurred.

Methods

This study was approved (approval number 21005) by the Ethics Committee on Experiments on Human Subjects of the University of Electro-Communications, Chofu, Tokyo, Japan. Informed consent was not obtained because the experiments did not involve human subjects.

The workplace where the cluster of infections occurred was a room where approximately 29 crew members stayed for approximately 1 hour in the morning and 1 hour in the afternoon to prepare for the next work process, while sharing same void with 23 desk staff members. The workplace volume is approximately 2,640 m³ and is divided by four partitions with a height of 1.8 m (Figure 1), so there was no face-to-face interaction and the possibility of droplet infection was relatively low. The distance between adjacent crews was approximately 2 m. Crew members work alone and are not required to communicate with each other, so communication between crew members were kept to a minimum. In the next work process, the crew worked at individual workstations that were far apart, so there was no physical contact.

Fig. 1. The layout of the workplace and the location of CO₂ sensors

The workplace is located on the second floor of a three-story building that was built in the 1950s and has been repeatedly remodeled. This workplace’s building is not equipped with a mechanical ventilation system able to ensure constant indoor air exchange rates. Therefore, the workplace is naturally ventilated through the opening of the available door and windows, and the timing and frequency of the opening/closing of a door and windows are generally decided by the administrators and are not scheduled. This can be partly related to the era of building construction (the 1950s), when the implementation of mechanical heating and ventilation systems was not contemplated yet in construction planning. Although hand disinfection and masks were strictly enforced during the cluster outbreak, only the doorway to another building was open for ventilation. This suggests that the risk of aerosol transmission was higher than that of contact or droplet transmission.

The COVID-19 cluster occurred in 2021. On weekdays, a total of 52 employees worked on this floor; 29 crew members performed preparatory work for the next work step at the workbench in the red-dotted frame in Figure 1. The other 23 desk staff members worked outside the red-dotted frame but shared the same void. One crew member was the index case, and 10 of the 29 crew members later tested positive for COVID-19 according to the polymerase chain reaction (PCR) testing. Additionally, one manager and one crew member also tested positive. The 29 crew members stayed in the room at once for approximately 1 hour in the morning. In the afternoon, the number of people simultaneously in the room was <29 at all times. Employees only worked together for approximately 1 hour in the morning and the afternoon; thus, they did not interact as a group, and the opportunity for mutual infection was limited. Despite the short time spent in the compartment, the relatively high infection rate among the crew members suggests that the infection spread due to causes confined to the compartment. A test of the difference in infection rates using a normal approximation by logit transformation of the binomial distribution yielded a p-value of 0.02 for the null hypothesis of no difference in infection rates. However, the location information of infected and uninfected persons is unknown, as the operator of this workplace did not provide it.

The compartmental indoor air ventilation was experimentally investigated based on the concentration decay tracer gas method²³⁾ using the CO₂ tracer gas sublimated from dry ice, which allows for quick and inexpensive application to a relatively wide range of sites. Although the continuous dose method and constant concentration method are also used for the tracer gas method, the concentration decay method was judged more suitable when using dry ice, for which it is difficult to control the flow rate generated. Compartmental transmission risk was indirectly evaluated using the CO₂ concentration as an alternative to the amount of exhaled air. We used dry ice as a CO₂ source, and eight CO₂ sensors were used to detect the changes in CO₂ concentrations in each compartment, as shown in Figure 1.

Non-dispersive infrared (NDIR) gas sensors were used as the CO₂ sensors. Eight TR-76Ui sensors (T&D Corporation, Nagano, Japan) were placed on the desks within the compartments where the employees worked (Figure 1). The numbers 1 to 8 in Figure 1 indicate the locations of each sensor. The TR-76Ui sensor can detect CO₂ concentrations from 0 to 9,999 ppm, with an accuracy of ±50 ppm (±5%).

The experiment was conducted from 10:30 am to 12:30 pm on November 28, 2021. Dry ice was crushed on the floor to vaporize CO₂, and the CO₂ concentration in the room was increased to approximately 3,000 ppm, which is sufficiently higher than the background level of outdoor CO₂, with no ventilation. Then, the decrease in CO₂ concentration owing to the ventilation was measured by each sensor from 11:17 am under ventilation Condition 1 without window opening. After 35 min, the ventilation condition was changed to Condition 2 with a window opening, and the decrease in CO₂ concentration attributed to ventilation was continuously measured from 11:52 am to 16 min. Condition 1 reproduced the situation at the time of the cluster occurrence, and Condition 2 represented an improved ventilation condition.

The ACH, representing the ventilation around the sensors, was estimated based on the time series change in CO₂ concentration under Condition 1. A transient mass balance model was used to solve the CO₂ concentration around the sensors. The stable mass balance of well-mixed air can be described as:

  

$$V\frac{d{C}_t}{dt}=M+\lambda C_0-\lambda C_t,$$(1)

Where C_t is the concentration of indoor pollutants at time t, M is the number of pollutants generated, λ is the air ventilation rate, and C₀ is the concentration in the absence of pollutant sources set to 400 ppm in this experiment. As a general solution to the first-order linear ordinary differential equation shown in Equation (1), the concentration of pollutants at a time is obtained by Seidel’s equation^24,25)

  

$$C_i=C_0+\left(C_s-C_0\right)e^{-\frac{Q}{V}\left(i-s\right)}+\left(1-e^{-\frac{Q}{V}\left(i-s\right)}\right)\frac{M}{Q},$$(2)

where Q is the outdoor air supply around the sensors [m³/h], V is the effective volume of the space around the sensors, s is the time at which the observation started, and λ=Q/V is the assumed air ventilation rate. When no pollutants are generated, i.e., M=0, Equation (2) can be transformed into:

  

$$ln\frac{C_i-C_0}{C_s-C_0}=-\frac{Q}{V}\left(i-s\right).$$(3)

This equation suggests that a decrease in the normalized concentration of CO₂ with the ventilation time (i – s) [h] without pollutant generation corresponds to the ACH in the space, which was estimated using the linear model of Q/V [/h] from Equation (3). Herein, the ACH was estimated, along with the point estimates, with a 95% confidence interval from the slope by assuming a linear equation with an intercept. The statistical software JMP Pro Ver. 16 (SAS Institute, Cary, NC, USA) was used for the regression analysis.

To estimate the ACH variation trend based on time series data analysis to clarify if the variation was significant compared to the observation errors, GLMM²⁶⁾ was used to analyze the effects of experimental conditions on ventilation and the differences and trends in ACH attributed to sensor location (ie, the inhibitory effects of inappropriate partitions on ventilation). The GLMM model is a mixed-effect model in which the ventilation time and interaction between ventilation time and sensor location are fixed effects and the sensor location is a random effect. The objective variable is the ratio of the increase in CO₂ concentration from the background to that at the beginning of the observation, and the model presented a natural logarithm as the link function.

Furthermore, the characteristics of the ventilation distribution were investigated based on the similarity of the ACH patterns at each sensor location. The dynamic time warping (DTW)^28,29) method was used to calculate the DTW distance of the ventilation patterns measured over time. DTW is an algorithm for measuring the similarity between two points of time-series data, which may vary in speed. Similarities in CO₂ variation can be detected using DTW, even if there are accelerations and decelerations during an observation. We used the statistical language R package “dtw” Ver. 1.22-3 (R Foundation for Statistical Computing, Vienna, Austria) to calculate the DTW distance.

Results

eFigure 1 shows time series variation of the observed CO₂ concentration during experiments. Figure 2 shows the estimated ACH values. Under Condition 1 (the same as during the COVID-19 cluster outbreak), the ACH was in the range of 0.408 to 1.178/hour (p<.001). Under Condition 2, where a window is opened to form a ventilation path, ACH improved to approximately 2/hour.

Fig. 2. Estimated compartmental air change per hour (ACH) according to the sensor. Error bars indicate a 95% confidence interval

The GLMM analyses for conditions 1 and 2 aptly demonstrated the observed data, with a coefficient of determination R² >0.96. The results of the fixed-effects tests are listed in Table 1, and Table 2 shows the estimated covariance parameters of the various effects. The results shown in Table 1 indicate that the interactions between ventilation time and sensor location were highly significant. Moreover, the decrease in CO₂ concentration with ventilation time for Conditions 1 and 2 was also highly significant. Random effects were not significant on their own for either Conditions 1 or 2, as shown in Table 2.

Table 1. Results of the test of fixed effects for each ventilation condition

Ventilation condition	Factor	Number of parameters	DOF of numerator	DOF of denominator	F-value	p-value
1	Ventilation time	1	1	264	4296.4964	<.001
	Interaction of Ventilation time and Sensor location	7	7	264	69.3716	<.001
2	Ventilation time	1	1	128	2610.0041	<.001
	Interaction of Ventilation time and Sensor location	7	7	128	4.8542	<.001

DOF, degrees of freedom.

Table 2. Covariance parameter estimates of random effects according to ventilation condition

Ventilation condition	Variance components	Estimate	Standard error	95% Lower limit	95% Upper limit	Wald p-value
1	Sensor	0.0092261	0.0049527	−0.000481	0.0189332	.062
	Residual error	0.0013827	0.0001204	0.0011742	0.0016526
	Sum	0.0106089	0.0049541	0.0050479	0.0348439
2	Sensor	0.0075253	0.0040821	−0.000476	0.0155261	.065
	Residual error	0.0020094	0.0002512	0.0015955	0.0026093
	Sum	0.0095347	0.0040890	0.0047692	0.0276991

Figure 3 shows the fitting results of the estimated GLMM to the observed values. The matrix of the DTW distance between each sensor data is shown in Figure 4. The DTW matrices were calculated for each ventilation condition. They were colored according to the distance, with a relatively high pattern similarity colored in green and low pattern similarity colored in red in a stepwise manner. The diagonal component was excluded because it refers to itself. The raw data of CO₂ concentrations were registered and are available from a repository in Microsoft Excel format.

Fig. 3. Fittings of estimated generalized linear mixed models to observed CO₂ concentrations

Fig. 4. DTW distance matrices for each ventilation condition. (A) Ventilation condition 1. (B) Ventilation condition 2.

Discussion

Under Condition 1, the ACH decreased as the distance from the doorway increased because the doorway was the only route to introduce outside air (Figure 2). At the farthest point, near sensor 8, ACH was less than 0.5/hour. According to the Japanese Ministry of Health, Labour and Welfare (MHLW), in workplaces without proper ventilation structures, ventilation should be provided at least twice an hour by opening windows and doorways⁸⁾. Furthermore, the probability of tuberculosis infection, for which aerosol transmission is the established route of infection, is markedly reduced in workplaces with an ACH of 2 or higher^29,30,31). The ACH during the COVID-19 cluster outbreak was probably much lower than this. Contrastingly, under Condition 2, the ACH trend by location was inversely proportional to that under ventilation Condition 1; its value increased as the distances from the doorway increased. This implies that the amount of outside air introduced by opening the window was dominant under ventilation condition 2.

Based on the findings in Table 1 and Table 2, it appears that the correlations between ventilation time and sensor location, as well as the reduction in CO₂ concentration over time for both Conditions 1 and 2, were extremely significant. These results suggest that the partitions installed in the room hindered proper ventilation and that the fluctuation in ACH was affected by the placement of the partitions and the direction of airflow.

As shown in Figure 3, the estimated GLMMs corroborated the observed values well, which indicates that the ACH estimates and their variations were reasonable. The observed values for sensors 1 and 8 in Condition 1, located at the room’s outermost periphery, showed a range of oscillations in the early stages of the measurement. Such a feature was not observed in Condition 2, indicating that the uniform stirring of the tracer gas may have been insufficient. The oscillation converged shortly after that, so its impact on ACH estimation is considered minimal.

The absolute values of the DTW distance between Conditions 1 and 2 could not be compared due to the differences in the number of available data (Figure 4). Therefore, we focus on the difference in room similarity distribution. Figure 4 shows that sensors 1, 7, and 8, near the edge of the room, presented less similarity in behavior with the sensors near the center of the room. The results suggest that the ventilation patterns of neighboring locations were similar. The range of pattern similarity near the center of the room was wider under Condition 2 in Figure 4B than under Condition 1 in Figure 4A. Although opening the windows in ventilation Condition 2 improved the ventilation of the entire workplace, the ventilation effect was less effective in reaching the center of the room because of the partitions. Therefore, it was speculated that ventilation could be increased by improving the airflow along the partitioned areas.

Although there is still a possibility of confounding variables owing to the relatively small number of infections in Japan since October 2021, the fact that no infections have been confirmed since the adoption of improved ventilation Condition 2 and the results of our previous study¹²⁾ indicate that the outbreaks of the infection clusters share the feature of an ACH of <2/hour. Additionally, where the ACH was improved to ≥2/hour, no evidence of a second infection cluster has been identified.

This index for ACH is consistent with the results of previous studies on tuberculosis and is considered highly valid^26,27,28). In large workplaces with intricate layouts and partitions, ventilation conditions become complex; the present study demonstrates that local monitoring and quantitative evaluation using sensor networks are effective. The results of another study by the authors^12,34,35) also confirmed that the ACH before the improvement was less than 2/hour at the COVID-19 clustering site, further supporting the results of this study.

Approximately 2,640 m³ of space was shared by 52 employees where the ACH in Condition 1 was in the range of 0.408 to 1.178/hour, which indicated, in particular, that there was stagnant air at the edges of the workplace where sensor 8 was placed. The ventilation rate per person in the sensor 8 area was estimated to be approximately 20 m³/hour, which is lower than the 30 m³/hour recommended by the Ministry of Health, Labour and Welfare for infection control¹⁸⁾, suggesting that the partitions generated relatively high-risk areas. This is consistent with the possibility that air convection caused by partitions may have contributed to the outbreak in smaller offices¹²⁾.

The building was not equipped with a mechanical ventilation system from its construction in the 1950s, and none was installed in several subsequent renovations. There are likely countless such buildings without mechanical ventilation that exist nonconforming in light of the Act on Maintenance of Sanitation in Buildings (enacted in 1970). Since there is no way for facility users to know whether modern renovated buildings are unventilated, they need education regarding ventilation risks that exist in older pre-1970 buildings.

In summary, for aerosol infection control, unnecessary partitions should be removed to not interfere with ventilation and even necessary partitions should be required to minimize their height and width. In conjunction with this, it may be helpful to use portable fans to promote localized ventilation if areas of air stagnation are anticipated.

According to the Act on Maintenance of Sanitation in Buildings in buildings, CO₂ concentration in a room should be measured at only one location in the room. However, by simultaneously measuring and analyzing multiple locations for each compartment, as was done in this study, it is possible to investigate the cause of COVID-19 clustering and prevent its recurrence. Additionally, the observed bias of CO₂ is more complex in rooms with larger sizes, complex geometries, and various uses. This agrees with previously reported results^12,22). Furthermore, these results will provide useful information for planning ventilation of public places, such as school classrooms³²⁾ or public buses³³⁾. In the future, real-time data from the CO₂ sensor network should be analyzed to identify compartments with increased risk to issue alerts at appropriate times. This would require combining CO₂ concentration and other sensor data, such as temperature, humidity, illumination, barometric pressure, and human detection. The authors have already reported a method of applying topological data analysis to multidimensional time-series data from many sensors²²⁾. In the future, we hope to develop a method for diagnosing anomalies by combining data integration with machine learning and deep learning.

Limitations

In this study, the workplace, which operates 365 days a year, was set to be temporarily unoccupied for this experiment, making it difficult to repeat the experiment, and only one experiment was performed. To avoid the influence of the small number of experiments, ACH estimation accuracy was statistically analyzed to confirm its significance. However, variations due to climate and seasons cannot be ruled out.

Additionally, this study employed the concentration decay method because of cost and time limitations. However, the constant dose method could effectively identify the source and visualize the aerosol flow. Furthermore, since this intervention experiment was conducted in only one building, the quantitative reproducibility in rooms of different sizes and structures is not ensured. However, our suggestions, such as opening windows, removing partitions, and confirming their local effects with multiple CO₂ sensors, would be effective in improving ventilation in any case.

Conclusions

We measured ACH using a CO₂ sensor network in a workplace where a cluster of COVID-19 infections actually occurred and investigated the adverse effects of inappropriate partitions and the details of the ventilation improvement effect. Simultaneously measuring multiple locations for each compartment could find the cause of COVID-19 clustering and prevent its recurrence. A statistical analysis using GLMM showed that the ACH in the room was biased to the position of the partitions and air flow root. Additionally, by examining the similarity of ventilation patterns using DTW, the distribution of ventilation patterns between the partition and the effect of ventilation conditions were evaluated. The results indicated that an ACH of <2/hour was favorable for preventing the COVID-19 clusters in this facility.

Sources of funding

This work was supported by a KAKENHI Grant (No. 21K19820) from JSPS.

Conflicts of interest

The authors declare that there are no conflicts of interest.

Data availability

The raw data of CO₂ concentrations are available from the following repository in Microsoft Excel format. DOI: 10.6084/m9.figshare.23822595

References

• 1.   Marinucci  M,  Pancani  L,  Aureli  N,  Riva  P. Online social connections as surrogates of face-to-face interactions: A longitudinal study under Covid-19 isolation. Comput Human Behav. 2022; 128: 107102.
• 2.  National Institute of Infectious Diseases of Japan. Route of transmission of a new coronavirus (SARSCoV-2). https://www.niid.go.jp/niid/ja/2019-ncov/2484-idsc/11053-covid19-78.html. Published March 28, 2022. Accessed January 8, 2020 (in Japanese).
• 3.   Jang  S,  Han  SH,  Rhee  JY. A cluster of coronavirus disease associated with fitness dance classes, South Korea. Emerg Infect Dis. 2020; 26(8): 1917-1920.
• 4.   Cai  J,  Sun  W,  Huang  J,  Gamber  M,  Wu  J,  He  G. Indirect virus transmission in a cluster of COVID-19 cases, Wenzhou, China, 2020. Emerg Infect Dis. 2020; 26(6): 1343-1345.
• 5.   Katelaris  AL,  Wells  J,  Clark  P, et al. Epidemiologic evidence for airborne transmission of SARS-CoV-2 during church singing, Australia, 2020. Emerg Infect Dis. 2021; 27(6): 1677-1680.
• 6.   Toyokawa  T,  Shimada  T,  Hayamizu  T, et al. Transmission of SARS-CoV-2 during a 2-h domestic flight to Okinawa, Japan, March 2020. Influenza Other Respir Viruses. 2022; 16(1): 63-71.
• 7.   Lu  J,  Gu  J,  Li  K, et al. COVID-19 outbreak associated with air conditioning in a restaurant, Guangzhou, China, 2020. Emerg Infect Dis. 2020; 26(7): 1628-1631.
• 8.   Khanh  NC,  Thai  PQ,  Quach  HL, et al. Transmission of SARS-CoV 2 during long-haul flight. Emerg Infect Dis. 2020; 26(11): 2617-2624.
• 9.   Azimi  P,  Keshavarz  Z,  Cedeno Laurent  JG,  Stephens  B,  Allen  JG. Mechanistic transmission modeling of COVID-19 on the Diamond Princess cruise ship demonstrates the importance of aerosol transmission. Proc Natl Acad Sci USA. 2021; 118(8): e2015482118.
• 10.   Jayaweera  M,  Perera  H,  Gunawardana  B,  Manatunge  J. Transmission of COVID-19 virus by droplets and aerosols: A critical review on the unresolved dichotomy. Environ Res. 2020; 188: 109819.
• 11.   Li  Y,  Qian  H,  Hang  J, et al. Probable airborne transmission of SARS-CoV-2 in a poorly ventilated restaurant. Build Environ. 2021; 196: 107788.
• 12.   Ishigaki  Y,  Kawauchi  Y,  Yokogawa  S,  Saito  A,  Kitamura  H,  Moritake  T. Ventilatory effects of excessive plastic sheeting on the formation of SARS-Cov-2 in a closed indoor environment. Environmental and Occupational Health Practice, 2022-0024. https://doi.org/10.1539/eohp.2022-0024-OA
• 13.   Peng  Z,  Jimenez  JL. Exhaled CO₂ as a COVID-19 infection risk proxy for different indoor environments and activities. Environ Sci Technol Lett. 2021; 8(5): 392-397.
• 14.   Schade  W,  Reimer  V,  Seipenbusch  M,  Willer  U. Experimental investigation of aerosol and CO₂ dispersion for evaluation of COVID-19 infection risk in a concert hall. Int J Environ Res Public Health. 2021; 18(6): 3037.
• 15.   Poydenot  F,  Abdourahamane  I,  Caplain  E, et al. Risk assessment for long and short range airborne transmission of SARS-CoV-2, indoors and outdoors, using carbon dioxide measurements. medRxiv. 2021:21256352. https://doi.org/10.1101/2021.05.04.21256352
• 16.   Schibuola  L,  Tambani  C. High energy efficiency ventilation to limit COVID-19 contagion in school environments. Energy Build. 2021; 240: 110882.
• 17.   Hou  D. Katal  A,  Wang  L. Bayesian calibration of using CO₂ sensors to assess ventilation conditions and associated COVID-19 airborne aerosol transmission risk in schools. medRxiv. 2021:21250791. https://doi.org/10.1101/2021.01.29.21250791
• 18.  Ministry of Health, Labour and Welfare (MHLW). Ventilation to improve “poorly ventilated enclosed spaces.” In: Shougyou shisetsu nado ni okeru “Kanki no warui misshuu kuukan” wo kaizen surutameno kanki ni tsuite [Commercial Facilities and others]. https://www.mhlw.go.jp/content/10900000/000616069.pdf. Published March 30, 2020. Accessed January 8, 2022.
• 19.  United States Centers for Disease Control and Prevention. Ventilation in buildings. https://www.cdc.gov/coronavirus/2019-ncov/community/ventilation.html. Published 2 June 2021. Accessed January 8, 2022.
• 20.   Shewhart  WA. Statistical Method from the Viewpoint of Quality Control. New York: Dover Publications Inc; 1939.
• 21.   Ishigaki  Y,  Kato  T. Proposal and implementation of a survey for prevention of aerosol infection - a case study in medical and elderly care facilities. Journal of Occupational Safety and Health. 2023; 16(2): 173-180.
• 22.   Endo  S,  Yokogawa  S. Analysis of the trends between indoor carbon dioxide concentration and plug-level electricity usage through topological data analysis. IEEE Sens J. 2022; 22(2): 1424-1434.
• 23.  International Organization for Standardization. Thermal performance of buildings and materials — Determination of specific airflow rate in buildings — Tracer gas dilution method. https://www.iso.org/standard/69817.html. Accessed August 17, 2023.
• 24.   Tarumi  H. Text of Architecture Environment: Design of Thermal and Air. Tokyo: Inoue Shoin Co., Ltd; 2007 (in Japanese).
• 25.   Ito  S,  Nishi  H. Estimation of the number of people under controlled ventilation using a CO₂ concentration sensor. IECON. 2012; 2012: 4834-4839.
• 26.   Dobson  AJ,  Barnett  AG. An Introduction to Generalized Linear Models. 4th ed. Boca Raton: CRC Press; 2018.
• 27.   Keogh  E,  Ratanamahatana  CA. Exact index of dynamic time warping. Knowl Inf Syst. 2005; 7(3): 358-386.
• 28.   Giorgino  T. Computing and visualizing dynamic time warping alignments in R: the dtw package. J Stat Softw. 2009; 31(7): 1-24.
• 29.   Menzies  D,  Fanning  A,  Yuan  L,  FitzGerald  JM; Canadian Collaborative Group in Nosocomial Transmission of TB. Hospital ventilation and risk for tuberculous infection in canadian health care workers. Ann Intern Med. 2000; 133(10): 779-789.
• 30.   Bloch  AB,  Orenstein  WA,  Ewing  WM, et al. Measles outbreak in a pediatric practice: airborne transmission in an office setting. Pediatrics. 1985; 75(4): 676-683.
• 31.   Toyota  M. Study on environmental factors of a tuberculosis outbreak in junior high schools. Kekkaku. 2003; 78: 11-16 (in Japanese).
• 32.   Di Gilio  A,  Palmisani  J,  Pulimeno  M, et al. CO₂ concentration monitoring inside educational buildings as a strategic tool to reduce the risk of Sars-CoV-2 airborne transmission. Environ Res. 2021; 202: 111560.
• 33.   Querol  X,  Alastuey  A,  Moreno  N, et al. How can ventilation be improved on public transportation buses? Insights from CO₂ measurements. Environ Res. 2022; 205: 112451.
• 34.   Kitamura  H,  Ishigaki  Y,  Ohashi  H,  Yokogawa  S. Ventilation improvement and evaluation of its effectiveness in a Japanese manufacturing factory. Sci Rep. 2022; 12(1): 17642.
• 35.   Ishigaki  Y,  Yokogawa  S,  Minamoto  Y,  Saito  A,  Kitamura  H,  Kawauchi  Y. Pilot evaluation of possible airborne transmission in a geriatric care facility using carbon dioxide tracer gas: a case study. JMIR Form Res. 2022; 6(12): e37587.

Appendix

eFigure 1. Time series variation of the observed CO₂ concentration during experiments: After increasing the indoor CO₂ concentration from 2500 to 3000 ppm, which is significantly higher than the outdoor background concentration, the windows and entrances were set to ventilation Condition 1; a gradual decrease in CO₂ concentration was observed in all locations of the room. Even after 34 min of ventilation, the CO₂ concentration only dropped to approximately 1,500-2,250 ppm. The variability in the CO₂ concentration also increased compared to that at the start of the observation period. Therefore, we changed the settings of the windows and entrances to ventilation Condition 2 to create a ventilation path and observed the changes in CO₂ concentration for 16 min. A slight improvement in the ventilation rate was observed. Additionally, CO₂ concentration variability decreased slightly. (The raw data is available at DOI: 10.6084/m9.figshare.23822595)