Content uploaded by Rik Heller
Author content
All content in this area was uploaded by Rik Heller on Nov 18, 2024
Content may be subject to copyright.
This documentation contains confidential information and is not for distribution. Disclosure outside of Wello.ai is strictly prohibited and is provided to understand
the method of validating claims to reduce employee absenteeism and hospital-acquired infections. Unauthorized sharing or distribution of this document is a
violation of confidentiality agreements and can result in legal action.
AI-Based Screening: A Targeted Approach to Reduce
Employee Absenteeism by Minimizing Spread Due to
Employee Presenteeism
Rik Heller1, Ralph Kaiser, Jonathan Shnitzer, Michael T. DiPaolo, Lauren Berk, Murray Cohen,
Thomas Taylor, Stacey DiSpigno
1 BioMed Department, Wello, Inc.
15960 Midway Rd, Addison, TX 75001, USA
1 Rik.heller@welloinc.com
This paper contains confidential and proprietary information
Abstract— This study explores the benefits of early contagion
detection and management to mitigate presenteeism, absenteeism,
and hospital-acquired infections in the healthcare sector.
This study presents the New Wellness Protocol, a novel
approach for identifying outlier temperatures with significantly
enhanced accuracy, resulting in a tenfold improvement in the
detection of true positives and negatives. Unlike traditional
methods, this protocol offers a dynamic determination tailored to
each employee, based on their historic temperatures.
By effectively discerning above-normal temperature events,
the New Wellness Protocol serves as a vastly superior indicator of
an employee’s contagiousness, thereby mitigating workplace
transmission risks. Importantly, its efficacy remains robust across
varying seasonal conditions. This heightened precision, coupled
with near-instantaneous detection capabilities, holds promise for
minimizing pathogen spread and consequent absenteeism.
This study analyzed over 28,000 temperature scans over two
years in a US healthcare facility. The researched algorithm
compared employee temperature deviations against the
conventional screening threshold of 100.4 °F, integrating this data
with COVID-19 Polymerase Chain Reaction test results. It found
that the program’s focus on above-normal temperatures
significantly outperforms traditional methods, with high positive
and negative predictive values.
In conventional screening methodologies, the primary
objective is to reduce the occurrence of false negatives and
enhance workplace safety. However, this investigation
acknowledges that simply reducing a predetermined high-
temperature threshold leads to a rise in false positives, potentially
causing disruptions for employers. Unlike other studies, the
evidence shows that this artificial intelligence method can reduce
these errors by an order of magnitude compared to other
screening methods, resulting in an additional five seconds
required to enter the workplace. The New Wellness Program
could transform the workplace health and safety landscape, with
a minor shift towards an identical but personalized and adaptive
screening process.
Keywords—Fever, Contagion Detection, Absenteeism,
Presenteeism, healthcare-associated infections, Non-Disruptive
Protocol, Asymptomatic Spread, Infection Prevention, Temperature
Screening, Workplace Health Management, Super-Spreader.
I. INTRODUCTION
Since the outbreak of SARS-CoV-2, attention to all aspects
of contagiousness and spread was focused on prevention. Early
in the pandemic, loss of life and disruption of daily life activities
seemed irreversible and had not been felt by the world since the
great influenza pandemic in 1919.
Although vaccination was shown to reduce the rates of
hospitalization and death, COVID-19 hospitalization rates are
still four times the rates of the most deleterious respiratory
influenza outbreaks. In general, senior citizens and their
healthcare staff remain vulnerable to COVID-19[1].
At present, effective methods to manage the spread are
viewed by institutions beyond healthcare as unbalanced and
disruptive. Universal masking, employee surveys, and
temperature screening were among the most effective methods,
and vaccination mandates were surprisingly disregarded as the
infection rates came down.
In 2009, the H1N1 pandemic gave rise to studies that
claimed that asymptomatic super-spreaders caused the spread of
the virus [2]. Interestingly, some studies observed that COVID-
19 outbreaks were claimed to be caused by the same [3].
Previous research has traditionally defined fever using a
single, constant temperature threshold (usually 100.4F), leading
to the belief that fever appears as a symptom with a delayed
onset spanning several days. In this study, the traditional
definition of fever resulted in approximately 90% of the
contagious employees being categorized as asymptomatic. Even
after overcoming the laws and controversies over biometrics like
facial recognition, prior research has not considered employees’
range of temperature, which is a factor over time [4]. Other
studies that have tested temperature-based methods commonly
used temperature measurements from the extremities,
particularly the wrists [5]. This strategy provides a reasonable
surrogate for evaluating normal body temperature, except during
infection. The hypothalamus increases body temperature during
infection by increasing blood flow (thus, the temperature) to the
delicate tissues of the core and head by reducing blood flow to
the extremities. This change creates a drop in temperature when
This documentation contains confidential information and is not for distribution. Disclosure outside of Wello.ai is strictly prohibited and is provided to understand
the method of validating claims to reduce employee absenteeism and hospital-acquired infections. Unauthorized sharing or distribution of this document is a
violation of confidentiality agreements and can result in legal action.
the body is in a febrile state, as shown by increased shivering
due to a higher fever state [6]. The AI method used in the study
considers that above-normal temperature events (ANTE) can be
symptomatic and asymptomatic. The ANTEs, as computed by
the AI, offer a surprisingly simple method for identifying bodily
infections across the stages from incubation to recovery.
Once more, this study determined that traditional screening
methods detected only around 10% of COVID-19-positive cases
compared to the AI New Wellness Protocol (NWP)
methodology. Furthermore, while isolation is appropriate for
those who are likely to be contagious, the National Institute of
Occupational Science and Health (NIOSH) experimentally
demonstrated how masking mitigated the viral spread
immensely [7]. Hence, the spread is mitigated when masking
isolates contagious employees.
This research extends beyond identifying and isolating
infectious employees. It reveals presenteeism as a cause of
frequent absences and underscores the risk of hospital-acquired
infections from community diseases.
II. HYPOTHESES
Hypothesis 1: An AI-based protocol that screens deviations
from employee temperature baselines outperforms current
methods, such as random testing, fixed temperature thresholds
(100.4 °F), or health questionnaires, in potentially showing
infectious diseases by reducing false negatives.
Hypothesis 2: An AI-based protocol that screens deviations
from employee temperature baselines ensures a much higher
negative predictive value when compared to current methods,
such as random testing, fixed temperature thresholds (100.4 °F),
and health questionnaires.
Hypothesis 3: Infectious employees serve as vectors for
disease transmission within the workplace. Using an AI-based
protocol to identify these employees and providing them with
NIOSH N95 masks can reduce the disease burden to 95% while
masking only 3% of the total workforce.
III. GLOSSARY
A glossary of terms ensures clarity and enhances readers’
understanding of the key concepts discussed in this paper. These
terms are fundamental to discussing and analyzing the NWP and
its application. The following definitions provide a quick
reference and aid in the comprehension of the specialized
terminology used throughout this paper:
ANTE (Above-Normal Temperature Event): This is a
significant deviation from an employee’s typical body
temperature. The ANTEs are crucial for showing early signs of
potential health concerns, particularly concerning infectious
diseases, where an employee’s temperature may deviate from
the norm but does not reach the Conventional Screening
Threshold.
CST (Conventional Screening Threshold): CST is the
commonly accepted benchmark in clinical diagnosis for
infection, typically set at 100.4°F (38°C).
NWP (New Wellness Protocol): NWP is an advanced
health monitoring approach based on personalized temperature
analysis. Unlike traditional methods, it uses employee
temperature baselines and deviations (ANTEs) to demonstrate
potential health risks more accurately, thereby enhancing the
effectiveness of temperature-based health screening.
Fever: Fever, as referred to in this study, is a fixed
temperature measurement, among other factors that healthcare
professionals use to diagnose infections. This study does not
invalidate the use of fixed temperatures in clinical diagnosis.
Fever and elevated temperature are among many factors
involved in clinical diagnosis in healthcare workers (HCWs).
IV. BACKGROUND: THE DYNAMICS OF WORKPLACE
CONTAGION AND CONTROL MEASURES
A. The Cause of Epidemics
The spread of infection causes outbreaks through casual
contact. Presenteeism causes the spread of infections in the
workplace [8]. Additionally, research indicates that a substantial
portion of HCWs, ranging from 50% to 90%, have reported
experiencing notable symptoms of infection in the workplace
[9]. Presenteeism is common at approximately 35% or more in
facilities outside of healthcare [10]. When studies show that
HCWs dealing with patients can walk up to five miles in a single
shift, presenteeism not only increases the risk of hospital-
acquired infections (HAIs) in patients but also raises concerns
about the numerous interactions between co-workers.
Interestingly, when the infectious patients were isolated using
negative differential room pressure, their ability to spread
infection was negligible [11]. Additionally, the staff-to-patient
ratio in the hospital on a single day was 10 HCWs per patient.
B. Disruptive Methods of Screening & Isolation
The current protocols for preventing the spread of disease in
the workplace are disruptive, if not controversial, in daily
operations. Such protocols include universal masking, require
100% rapid testing of all employees, screening by survey, and
fixed temperature scanning. Rapid tests are, for now, limited to
a single infection. With the advancing rate of test discovery, a
wide range of diseases may be detectable by single antigenic
tests or Polymerase Chain Reaction (PCR) tests. With a wide-
spectrum test, the employer would bear the costs of a proctor for
the test, one non-reusable test per employee, and the added time
required per employee to administer.
C. Lytic Cycle, Viral Shedding, and Detection Challenges in
Infectious Disease Control
Elevated temperatures are the immediate result of the body’s
detection of viral infections, such as influenza, RSV, and
Coronaviruses, caused by a lytic replication cycle [12]. In this
cycle, the virus hijacks the replication apparatus of the host cell
to replicate the virus, lyses (bursts) the host cell, and releases
hundreds of new virions. Then, the body detects the virus and
remnants of broken host cells, initiating several mechanisms that
elevate temperature and a systemic defense that has evolved for
over 600 million years to stop the virus from replicating [13].
Viral shedding is a direct function of the growing respiratory
virus concentration in the bloodstream [14] and is transmitted
through exhalation.
This documentation contains confidential information and is not for distribution. Disclosure outside of Wello.ai is strictly prohibited and is provided to understand
the method of validating claims to reduce employee absenteeism and hospital-acquired infections. Unauthorized sharing or distribution of this document is a
violation of confidentiality agreements and can result in legal action.
This lytic cycle begins at the start of incubation, as does
shedding, making the infection detectable days before high fixed
temperatures occur.
To further outline the importance of this:
D. Elevated Body Temperature and Viral Shedding
The correlation between higher temperatures and increased
viral shedding suggests that higher temperatures <from normal>
could be more infectious [15].
E. Positive PCR Tests Without Elevated Temperature
Often, employees test positive for a virus via serum tests
without symptoms, and their shedding, if any, would have little
to no impact on the spread of the virus.
F. Hurdles in Temperature-Based Screening in Large
Populations
A Harvard University study highlights practical challenges
in implementing employee temperature self-monitoring as a
preventive strategy for large populations [16].
• Perceived Effectiveness: Temperature screening during
the COVID-19 pandemic would be akin to a “security
theater.” Instead of mitigating the spread of the virus, it
offers a false sense of security.
• Quantitative vs. Qualitative Monitoring: As a
compromise, the study recommends qualitative
measures, such as self-reported symptoms, over self-
monitoring fixed temperatures.
• Compliance in Employee Monitoring: Only 30.4% of the
participants consistently recorded their daily
temperatures.
• Financial and Logistical Burdens: The cost of supplying
thermometers to participants was over USD 18,200.
• Equipment Availability: Many participants, especially
students, did not own thermometers.
Oral thermometry measured for 3 min is the gold standard
against which most minimally invasive thermometers are rated.
However, participants’ eating or drinking 15–30 minutes before
the experiment could have affected the results [17].
Alternatively, the use of 15-second tests led to reduced accuracy.
Finally, proper placement in the sublingual pocket for 3 minutes
was optimal.
G. The Misconception of “Normal” Body Temperature
The clinical definition of “normal” body temperature is
challenged here:
• Broad Range of Normal Temperatures: Contrary to over
a century of study and recognized in research over recent
decades, 98.6 °F (37.0°C) is not the average body
temperature across human populations. The FDA, too,
acknowledges that “normal” body temperature ranges
from 97 °F (36.1°C) to 99 °F (37.2°C).
• Recent Findings on Average Body Temperature: Recent
studies indicate that the average body temperature may
be lower than 98.6°F, with data suggesting a mean closer
to 97.9°F (36.6°C) and a standard deviation across that
population of ~1.8°F (1°C) [18].
• Factors Influencing Body Temperature: An employee’s
body temperature varies due to multiple factors,
including sex, age, ethnicity, environmental conditions,
and time of day.
Fig. 1. Temperature Range by Individual
H. The Significance of Masking Infectious Employees
Movert et al. illustrated that respiratory droplets and droplet
nuclei emitted through everyday actions such as breathing,
speaking, singing, and coughing can travel distances of 26 ft
(8m). This length is significantly greater than the standard social
distancing recommendation of 6 ft (1.8m). Figure 2a illustrates
this air propagation
Furthermore, Figure 2b shows that when employees
maintain the advised social distance, the potential for infection
remains because of nuclear droplets, which may carry suspended
low-density lyophilized virions that spread beyond this distance.
Compounding factors such as drafts, indoor thermoclines, and
lower humidity can further increase the risk of spreading the
virus.
Masks on the contagious surface serve as a significant barrier,
reducing the distance between the droplets and droplet nuclei
and decreasing the spread risk. This characteristic is shown in
the above-mentioned figures and the graphic example below,
where 100 people could be in a room together, such as a facility
cafeteria.
This documentation contains confidential information and is not for distribution. Disclosure outside of Wello.ai is strictly prohibited and is provided to understand
the method of validating claims to reduce employee absenteeism and hospital-acquired infections. Unauthorized sharing or distribution of this document is a
violation of confidentiality agreements and can result in legal action.
Fig. 2. Potential for Spread of Infection.
V. METHODS
A. Experiment Design
1) Algorithm Overview
The study assessed two algorithms for identifying ANTEs,
selecting the weather decorrelation algorithm Figure 3 for its
superior accuracy, as indicated by the findings:
1. Weather Decorrelated Algorithm: Adjusts temperature
readings for weather, ensuring the data reflects an employee’s
health status while considering local weather influences.
2. Standard Algorithm: Works without considering
weather variations.
The analysis uses the Weather Decorrelated Algorithm.
Fig. 3. Weather Decorrelated Algorithm.
This approach considers the uniformity of employee
physiological factors, such as gender, age, race, and the time of
day, across repeated temperature check-ins. These parameters
were excluded in the AI adjustments for privacy, leaving only a
subject’s learned reaction to weather in the AI models. Figure 4
shows an example of weather adjustments (blue dots) for an
employee’s temperature.
B. Normalization of Recorded Temperatures
This study normalized all recorded temperatures by
expressing them as deviations σ (sigma) from employee mean
temperatures in terms of standard deviations. The deviation (σ)
per employee ranged from −3.1 to +2.9.
This first step was to create a baseline from which the
outliers could be identified. By standardizing the temperatures
this way, each employee’s measures were compared against
their own baselines. Subsequently, by establishing a deviation
threshold, it became easier to identify outliers in comparison to
each employee’s baseline.
This documentation contains confidential information and is not for distribution. Disclosure outside of Wello.ai is strictly prohibited and is provided to understand
the method of validating claims to reduce employee absenteeism and hospital-acquired infections. Unauthorized sharing or distribution of this document is a
violation of confidentiality agreements and can result in legal action.
C. Validation Method for Hypothesis 1
The study applied the following methods to verify the
algorithm’s efficacy:
1) Direct Comparison Method
• Random Analysis Comparison: Our algorithm was
employed to detect outliers within our population,
followed by a random selection of an equivalent number
of outliers from the same population. Subsequently, we
assessed the algorithm’s success rate in identifying
employees with illnesses, as well as the success rate of
random selection in the same regard. Success rates were
determined by cross-referencing against confirmed
positive PCR tests.
The ratio between those two “success rates,” quantifies the
effectiveness of the method compared with random choice.
2) Indirect Comparison Methods
• Attendance Analysis Post-Outlier Identification: The
paper checked attendance patterns after identifying
outliers by the algorithm, assuming that employees with
elevated temperatures would show a higher absence rate
due to illness.
• Longitudinal Attendance Tracking: A visualization of
attendance trends and potential illness onset was tracked
for ten days following their outlier event (designated as
"Day Zero").
D. Validation Method for Hypothesis 2
This method involved analysis of the sensitivity, specificity,
positive predictive value (PPV), and negative predictive value
(NPV) of ANTEs. It was employed to understand how well the
method distinguished between actual cases of illness and
incorrectly flagged employees.
• Statistical Modeling: Data on regional COVID-19
prevalence were integrated to simulate a normal
distribution curve reflecting the characteristics of the
confirmed sick population. The study used regional
prevalence data to scale up this curve to its proportionally
expected size.
• Analysis of PCR-Confirmed Outliers: The study focused
on outliers confirmed positive by PCR testing to analyze
their average deviations and show the characteristic
pattern of temperature anomalies in the sick population.
This analysis revealed typical patterns of temperature
deviation in the ill population. However, the precise size
of the sick population curve remained unknown,
prompting us to integrate regional prevalence data for
accurate scaling of the curve.
• Calculation of Standard Metrics: Any event with an
ANTE but no positive PCR test result was subtracted
from the total population in the model. The PPV, NPV,
sensitivity, and specificity were calculated using
established statistical methods to gauge the technique’s
performance in identifying ANTEs.
E. Instrumentation for Temperature Data Collection
The instrument for temperature data collection was the
welloStationX (WSX), a non-contact infrared thermometer
(NCIT) with details that include:
• FDA Class II fully automated self-service thermometer:
FDA unique product code FLL, indicating a registered
clinical electronic thermometer.
• Automated facial detection: Equipped with a camera and
facial detection technology to help users align their faces
and guide them to align themselves correctly for the
temperature scan when facing the LCD screen using both
visual and audio aids for precision.
• Automatic scan initiation: The temperature scanning
process begins automatically when the user is correctly
positioned.
• Data Upload and Retrieval: In online mode, WSX
uploads the collected temperature data to the Wello
Cloud database upon completion of the temperature
scan.
F. Study Setting and Data Collection
The study reviewed a high-test facility for this study [19].
This US senior living and hospital establishment has reduced
mortality rates, as found through several employee tests. Thus,
it facilitated early detection of COVID-19(24). The employee
temperatures were assessed daily. This facility adopts a
multifaceted approach to screening for potential COVID-19
cases, combining temperature screening of all employees with
other criteria and COVID-19 tests to assess residents and staff.
G. Ethical Approval and Informed Consent
The analysis in this study is retrospective, with employee
identities and actual locations irretrievably encrypted. All
measurement data are of employees and measurements are
required for entering the facility. No patients or visitors were
permitted to use the equipment.
H. Hospital Temperature Screening Process
• The primary method of temperature screening employed
at this hospital was the WSX.
• Data from the WSX was securely uploaded and collected
for this study.
• For employees not screened by WSX, manual
temperature screening was conducted using NCITs.
However, this study did not include the data from these
manual screenings.
Employees had to record their temperature before starting
their shift and could optionally scan temperatures multiple times
throughout the day.
I. Health Monitoring and Survey Data
• The hospital routinely conducted health surveys to assess
the status of employees.
• Employees feeling unwell were advised to stay home or
leave the facility and had to schedule independent
This documentation contains confidential information and is not for distribution. Disclosure outside of Wello.ai is strictly prohibited and is provided to understand
the method of validating claims to reduce employee absenteeism and hospital-acquired infections. Unauthorized sharing or distribution of this document is a
violation of confidentiality agreements and can result in legal action.
COVID-19 testing unless testing was conducted on-site.
Testing at the site was conducted using PCR.
J. COVID-19 Testing and Return-to-Work Protocol
• In cases of fixed high temperatures (above 100.4°F), the
employees underwent a waiting period of 15 min in
designated areas for re-scanning. They were instructed to
leave the premises and notify the Employee Health and
Safety manager if the temperature remained high.
• A mandatory PCR test was conducted before these
employees left the facility.
• The return-to-work criteria included a negative test result
and being symptom-free for one week.
K. Data Analysis Approach
The study analyzed the temperature data using an algorithm
designed to detect ANTEs. This algorithm was not actively used
in hospitals during the data collection period. Positive COVID-
19 test results were incorporated with the corresponding
employee data to analyze the efficacy of the ANTE detection
approach.
VI. RESULTS
The first phase of the trial involved analyzing the core body
temperature scans of 336 employees at a healthcare facility in
the northern United States. The scans were performed during
daily check-ins from October 1, 2022, to March 31, 2023,
accumulating 27,994 employee scanning events. The study
encrypted and obscured employee IDs to protect privacy while
retaining their unique identifiers for analytical purposes.
Furthermore, it excluded IDs linked to fewer than 15 scan events
to support statistical relevance.
In the subsequent phase of this study, our focus shifted to the
analysis of 500 positive COVID-19 PCR tests conducted on a
cohort of 91 employees between September 2020 and March
2023. Given our specific interest in these 91 employees, we
systematically excluded all data not about them while retaining
relevant information with extended date ranges. The
comprehensive dataset for this subsequent phase encompasses
28,309 scan events.
A. Outlier Identification via Employee Threshold
Determination
By setting up individualized temperature thresholds and
expressing them as multiples of the standard deviation, the study
categorized the temperature readings into normal range and
ANTEs. It defined ANTE as any temperature measurement
exceeding 1.7 standard deviations from an employee’s baseline
temperature below 100.4 °F. Notably, the selection of 1.7 as the
threshold value was optimized to achieve a cost-effective
balance. This strategy minimizes the risk of disease transmission
while ensuring a seamless integration into the routine operations
of a healthcare facility. Further elaboration on this optimization
process is provided in subsequent sections of this paper.
Figures 4 and 5 show results from the initial data collection
between October 1, 2022, and March 31, 2023. To enhance
clarity and avoid overwhelming information in the chart, we did
not include the full dataset from September 2020 to March.
Figure 4 shows the categorization and quantification of all
temperature points across different ranges: non-outliers, outliers
within 98.6 °F to 100.4 °F, outliers below 98.6 °F, and outliers
above 100.4 °F.
Fig. 4. Individualized Temperature Deviations.
Figure 5 visualizes the temperature outliers when mapped
back onto a body temperature scale with the same temperature
range categories described above. The number of outliers for
each range was as follows:
• Below 98.6°F = 767 outliers
• 98.6°F to 100.4°F = 136 outliers
• Above 100.4°F = 9 outliers
Fig. 5. Temperature Mapping of Phase One Employees (n=336).
B. Hypothesis 1: Testing for Effectiveness
1) Part A: 1.7 Standard Deviations vs Random
This section evaluates the algorithm’s effectiveness in
finding outliers corresponding to PCR-confirmed COVID-19
cases within a ±10-day window, labeling such instances as
“matches.” Although the algorithm effectively flags potential
COVID cases, elevated temperatures might indicate other health
conditions.
Fig. 6. Temperature Scan History.
This documentation contains confidential information and is not for distribution. Disclosure outside of Wello.ai is strictly prohibited and is provided to understand
the method of validating claims to reduce employee absenteeism and hospital-acquired infections. Unauthorized sharing or distribution of this document is a
violation of confidentiality agreements and can result in legal action.
Figure 6 depicts the temperature history of an individual
outlier identified by our method that aligns closely with a near
PCR-confirmed COVID case. The algorithm correctly detected
32 PCR-confirmed cases of infection from a subset of 136 cases
(See limitations section) using temperature data alone. The study
design considered attendance biases caused by external
laboratory testing requirements for symptomatic employees,
affecting the availability of temperature data on the day of PCR
diagnosis.
In addition, an equivalent number of outliers, as
identified by our algorithm, were randomly selected from our
overall population, allowing for a comparison of the success
rates between a random selection approach and our proposed
method. After 1000 iterations, our algorithm demonstrated a
4.51 to 1 advantage in accurately identifying confirmed cases.
2) Part B: Correlation Between Temperature Anomalies
and Attendance Patterns
Analysis of attendance patterns following temperature
anomaly detection revealed a correlation with decreased
workplace presence, suggesting a link between detected
temperature deviations and health-related absences. Positive and
negative controls were established by monitoring attendance
patterns following PCR-confirmed cases and randomly selected
non-positive scan events, respectively. The former served to
validate the efficacy of the implemented measures, while the
latter provided a baseline for comparison.
Regular attendance cycles, including expected fluctuations
due to weekly rest days, were considered in the analysis to
isolate the impact of health-related absences from regular
attendance behaviors. Day Zero denotes the date an employee
received a positive PCR test. Positive Control: Workplace
Absence Trends Post-PCR Confirmation
The results in Figure 7 show a marked decrease in attendance
from days +1 to +4 following a PCR-confirmed positive test, in
line with the enforced workplace policy that mandates a five-
day absence for employees testing positive. This trend may also
reflect instances where employees are too ill to attend work.
Fig. 7. Positive PCR Test Results on Cumulative Attendance.
3) Negative Controls: Analysis of Attendance Patterns in
Non-Positive Scans
This section focuses on the natural attendance behavior of
the study’s population without ANTE and positive PCR events.
Each control group was established by randomly selecting 500
scan events from the phase two dataset of 91 employees. Figures
8 and 9 illustrate the typical attendance patterns of employees in
scenarios devoid of specific events or health interventions.
Fig. 8. Healthy Negative Control I.
Fig. 9. Healthy Negative Control II.
4) Experimental Data: Impact of ANTEs only on Employee
Attendance Patterns
This section analyses the attendance data concerning
employees flagged by only ANTEs. Figure 10 shows variations
in attendance patterns after the identification of the ANTEs.
Specifically, a trend of reduced attendance post ANTEs, which
differed from regular attendance behavior, was observed among
these employees.
Fig. 10. Attendance Data Concerning Employees Flagged by Only ANTEs.
C. Hypothesis 2: Estimating PPV, NPV, Sensitivity, and
Specificity
1) Identification and Normality Assessment of the Overall
Population
This section assesses the normality of the temperature data
collected from the population. Two methods are used for this
purpose.
1. Frequency Diagram Analysis:
This documentation contains confidential information and is not for distribution. Disclosure outside of Wello.ai is strictly prohibited and is provided to understand
the method of validating claims to reduce employee absenteeism and hospital-acquired infections. Unauthorized sharing or distribution of this document is a
violation of confidentiality agreements and can result in legal action.
a. The frequency diagram shown in Figure 11a was
created by plotting all temperatures recorded during the study
period.
b. The resulting diagram showed a distribution that
visually aligns with a normal curve.
Fig. 11. Frequency Diagram for Overall Study Population.
2. Q-Q Plot Verification:
a. A quantile–quantile (Q-Q) plot was employed as an
independent method to verify the distribution of the study’s
dataset [20].
Figure 11b shows a close alignment with a 45-degree
straight line, indicating a close-to-normal distribution.
2) Identification and Normality Assessment of the Sick
Population
In this section, the study focuses on distinguishing the sick
population within the sample by aligning the algorithm’s
predictions with the positive PCR test results. The process
involved the following steps.
1. Selection of Sick Population Data The study selected
scan events that consistently matched the algorithm’s
predictions and positive PCR test results.
2. Normalization Method Given that the study’s
methodology filtered out all data points below 1.7 standard
deviations (σ), we employed a mirroring approach to augment
the lower tail of the distribution, thereby normalizing the
representation of the diseased population.
3. Assessing Normality. A Shapiro-Wilk normality test
was conducted, resulting in a p-value of 0.861. This result
suggests non-rejection of the null hypothesis, showing that the
data of the sick population adhere to a normal distribution. A (Q-
Q) plot analysis was also examined in Figure 12, which further
supports the normal distribution of the dataset (27).
Fig. 12. Quantile–Quantile Plot for Sick Population.
Graphical analysis: Figure 13 shows a near normal
distribution of the sick population data.
Figure 13: Frequency Plot for Sick Population.
3) Performance Estimation and Curve Simulation for the
Sick Population
In this phase, the study estimated the performance
characteristics of the study’s method for an identified group of
sick population. The process involved the following steps.
1. Estimation of Mean and Standard Deviation
a. The study calculated the mean value of the temperature
variance from each employee’s baseline in the sick population.
Their mean was found to be 1.7.
b. The standard deviation associated with this variance
was calculated to be 0.27.
2. Simulation of the Sick Population Curve
a. Based on the calculated mean and standard deviation
and under the observation of data being normally distributed, the
This documentation contains confidential information and is not for distribution. Disclosure outside of Wello.ai is strictly prohibited and is provided to understand
the method of validating claims to reduce employee absenteeism and hospital-acquired infections. Unauthorized sharing or distribution of this document is a
violation of confidentiality agreements and can result in legal action.
study simulated the distribution curve for the sick population
using regional prevalence data.
b. The simulation incorporated the regional prevalence of
COVID-19 cases, aligning with a proportional representation of
140 cases throughout the study period within our specified
population.
After that, the study focused on simulating sickness events
in the overall population. This method involved a normally
distributed random selection of scan events from the overall
population that matched the curve of the sick population, which
were then selected as the simulated positive events. These events
were subtracted from the overall population to segregate the
healthy and sick populations.
A cutoff limit was set to help distinguish between healthy
and sick employees, as shown in Figure 14. Estimated counts of
true positives (actual sick employees correctly identified by the
system); true negatives (signifying healthy employees
accurately identified); false positives (healthy employees
mistakenly tagged as sick); and false negatives (sick employees
inaccurately categorized as healthy).
Figure 14. Cutoff for True and False Positives and Negatives.
D. Hypothesis 3: Analysis of Propagation Patterns
After plotting each participant’s ANTE against a combined
timeline, the study analyzed the potential transmission patterns.
Figure 15 delineates employee trajectories, where each row
corresponds to a distinct identifier (ID). Within these lanes, the
plotted curves highlight five symmetrically selected data points
centered around the deviation peak.
Figure 15. Propagation Timeline.
The gradation of color across the 14 employees represents
the extent of deviation from the employee’s established baseline
temperature. The color intensity shifted from a lighter shade to
deep red as the deviation increased, with the most saturated red
indicating an outlier that exceeded four standard deviations from
the mean baseline.
VII. DISCUSSION
A. Summary: Key Findings
This study analyzed employee temperature variations to
detect potential communicable diseases within a community.
The central aim of this study was to confirm the efficiency of
ANTEs as indicators of illness and to evaluate their predictive
value for disease detection compared to traditional methods. The
key data-driven findings are as follows:
• Temperature Baseline Establishment: The study
established employee temperature baselines for
approximately 300 employees. The baseline
temperatures varied significantly between 95.9 °F and
98.2 °F, showing the inaccuracies of fixed temperature
screening.
• Identification of ANTEs: The study detected ANTEs in
1046 cases where the 91 phase two employees showed
temperatures beyond 1.7 standard deviations from their
established baselines. These events, triggered below
100.4°F, correlated strongly with reported symptoms or
confirmed illness.
• Comparison with Random Testing: When simulating
ANTE-based detection against random testing, results
from 500 simulations involving randomly selected
employees showed a 4.51: 1 separation gained over
random simulation. This suggests a significantly higher
likelihood of detecting infections using ANTE-based
detection.
This documentation contains confidential information and is not for distribution. Disclosure outside of Wello.ai is strictly prohibited and is provided to understand
the method of validating claims to reduce employee absenteeism and hospital-acquired infections. Unauthorized sharing or distribution of this document is a
violation of confidentiality agreements and can result in legal action.
• Positive Predictive Value and Negative Predictive
Value: The ANTE-based detection method achieved a
PPV of 14.9% and NPV >99%. These values confirmed
the method’s precision for identifying infections with
minimal false positives. It results in the low-cost action
of wearing a facemask when flagged, thereby reducing
the risk of the potential excessive cost associated with the
spread of the disease. The protocol was perfected to
reduce false negatives, crucial for preventing disease
spread and reducing absenteeism, with a slight increase
of less than 3% in false positives.
• Correlation with PCR Test Results: The study observed
a significant correlation between ANTE occurrence and
positive PCR test results for COVID-19. Out of 136
ANTEs, 32 were positive for COVID-19 PCR results
(See limitations section).
• Observations of Illness Clusters: Notable clusters of
illness were identified using ANTEs, showing the
likelihood of the disease spreading within the workplace.
B. Theoretical and Practical Implications
In the segments below, the broader implications of the
findings are discussed.
• Presenteeism:
o While presenteeism may appear to support
productivity superficially, it often reduces work
efficiency and quality, as these workers contribute to
the spread of infectious diseases. This trend led to a
wider range of outbreaks and increased absenteeism.
Employees working while ill could be less capable of
performing their best, which can lead to errors,
decreased output, and long-term productivity losses.
Implementing advanced screening measures allows
organizations to mitigate the effects of presenteeism
by showing and addressing potential infections before
they manifest visible symptoms, leading to lower
disease spread and increased productivity.
• Absenteeism and its Implications:
o Absenteeism has significant implications for both
employee and organizational health. It is recognized
as a critical determinant of productivity and national
development as it impacts both the national and global
economy [21]. The consequences are a decrease in
production and quality of service. In addition, there is
an added load on colleagues and stress on the
relationships between employers and employees [21].
o A comprehensive policy to manage absenteeism
should involve multiple strategies, like motivating
administrators/employers to increase organization
commitment and to measure and manage
appropriately a healthy work environment and healthy
human relations” [21].
o The early detection of illness through advanced
screening can reduce the negative impact of
absenteeism. Organizations can support productivity,
relieve strain on other employees, and foster healthier
and more stable working environments.
• Prevention of Worker-to-Patient HAIs:
o Regular screening of HCWs for early signs of
infection protects patients and ensures the health and
well-being of staff. By identifying and managing
potential infections among HCWs early, the risk of
HAI spreading to patients from sick workers can be
reduced significantly.
C. The New Wellness Protocol: A Proposed Methodology
The NWP is designed to enhance the effectiveness of health
screening in workplace and institutional settings to mitigate the
effects of presenteeism, absenteeism, and HAIs. This protocol
integrates AI-based temperature screening with robust data
analysis and practical health and safety measures. The NWP
method is outlined as follows:
1) Temperature Recording and Data Collection
• Individualized Screening: Each employee’s baseline
temperature was recorded using a clinical thermometer.
• Identification and Data Protection: Employee IDs or
other unique identifiers were used for data collection and
tracking. This strategy ensures the protection of personal
health information.
• Ambient Temperature Consideration: Ambient or room
temperature was also recorded, as it can influence the
body temperature when using NCITs.
2) Analysis of Temperature Variations
• Baseline Temperature Analysis: The protocol involved
analyzing employee temperature readings against each
employee’s established baseline and considering their
normal temperature fluctuations.
• Detecting ANTEs: An ANTE is appropriately flagged
and the employer is notified in real time.
3) Health and Safety Measures for ANTEs
• Mandatory Mask Requirement: Employees flagged with
an ANTE must wear a mask to prevent the potential
spread of infection.
• Optional Disease Testing: Besides wearing a mask, those
employees may opt for disease testing (e.g., COVID-19
or flu tests). Their choice depends on the resources and
requirements of the facility
4) Integration with Existing Health and Safety Processes
• Evaluation of Current Protocols: Organizations are
encouraged to evaluate their current health and safety
protocols to seamlessly integrate NWP, as it can be
adapted and customized to fit various organizational
structures.
5) Data Management and Privacy Compliance
• Secure Data Handling: All collected data, including
temperature readings and health safety actions, are
managed securely and adhere to privacy laws and
regulations.
This documentation contains confidential information and is not for distribution. Disclosure outside of Wello.ai is strictly prohibited and is provided to understand
the method of validating claims to reduce employee absenteeism and hospital-acquired infections. Unauthorized sharing or distribution of this document is a
violation of confidentiality agreements and can result in legal action.
• Data Analysis for Health Insights: The collected data
were analyzed to gain insights into health trends and aid
in proactive health management.
The study provides important insights into using
individualized temperature monitoring to detect potential
infectious diseases in workplace environments. However, it is
crucial to acknowledge the limitations of this study to gain a
comprehensive understanding of its scope and implications.
D. Limitations and Assumptions
• While the dataset accounts for environmental factors,
such as weather and ambient temperature, it does not
distinguish between ANTE caused by COVID-19 and
those resulting from other illnesses.
• Vaccination can influence the immune response and
potentially affect the progression of COVID-19, so
vaccination status is a factor to consider for each
employee in future research.
• Since WelloStationX operates by measuring the
forehead/canthus, glasses, and headwear may alter
temperature readings. These obstructions should be
considered to ensure the greater accuracy of true
negatives.
• Check-in after positive diagnosis: Thirty-three out of 500
employees checked back within five days after testing
positive. This observation raises concerns about the
consistency in handling infectious employees or
potential inconsistencies in PCR test logs, which need to
be resolved in future research.
• Insufficient Temperature Scans: One hundred twenty-
nine PCR analyses within the studied population were
excluded due to insufficient temperature scans (fewer
than 15 scans per employee). This strategy arises from
the challenge of providing a meaningful baseline with a
limited dataset of temperature measurements or ensuring
greater compliance with automated self-service testing.
• Another 103 PCR-positive events were excluded
because they were classified as “rogue tests,” where no
temperature data were available regarding the employee
30 days before and after the positive PCR event. This
exclusion was necessary to ensure the reliability of the
data and mitigate the impact of potential false-positive
results.
• Temperature Inconsistencies in PCR-Positive Events:
Ninety-nine employees who assessed positive for PCR
exhibited temperatures below their averages. This
finding could be because of the high sensitivity of PCR
tests, which can detect the virus even after symptomatic
resolution.
• The study assumed a 100% PPV for the PCR test results.
In reality, PCR tests for COVID-19 have lower PPV and
false positives. Future research should consider these
variations in PPV to improve diagnostic accuracy and
reliability.
E. Limited success rate on false negatives:
1. The use of antipyretics could potentially lower
temperatures, leading to false negatives.
2. In many cases, elevated temperature periods may have
been shorter than anticipated. Our reliance on temperature data
from days before and after the positive test stems from the fact
that a significant portion of employees tested did not have
temperature scans recorded on the exact date of testing.
F. Limited Success Rate on False Positives:
1. Other infections could also cause elevated body
temperatures, leading to false positives.
2. Personal behaviors, such as biking to work on a
specific day, could result in punctual anomalies.
G. Future Research
The promising results of this study showing ANTEs used in
a workplace set the groundwork for future research avenues.
These areas are crucial for enhancing the effectiveness of
temperature-based health monitoring systems, such as the NWP,
and addressing the current study’s limitations.
• Comparative Effectiveness of NWP vs. Traditional
Methods: Future studies should compare the
effectiveness of NWP, its focus on ANTEs, against CST
and health questionnaire-based screening.
• Broader Population Studies: Future studies using the
NWP should encompass a more diverse range of
employees across multiple facilities. Research in varied
demographic, geographic, and environmental settings
would yield a more comprehensive understanding of the
effectiveness and adaptability of NWPs.
• Technology Adjustments for Facial Coverings:
Investigating advanced detection technologies that
account for the impact of facial coverings and glasses on
temperature readings to improve accuracy.
• Environmental Acclimation: Investigating AI controls
for employee responses to outdoor weather. Future
evaluations using this AI technique may assist in
predicting body response to weather conditions.
• Integration with Other Health Information: Combining
the NWP with other personal health monitors, such as
smartwatches, could create a more comprehensive
screening approach. Research in this area will provide
further input regarding the study’s AI for early disease
detection.
VIII. CONCLUSIONS
A. Brief Summary of Results and Discussion
The study introduced the NWP, which focused on finding
ANTEs as a more nuanced approach for detecting potential
illnesses by requiring non-disruptive and immediate follow-up
action. This screening method is significantly more effective
than traditional CSTs, which rely on fixed temperature
benchmarks such as 100.4°F (38 °C) and health questionnaires.
This documentation contains confidential information and is not for distribution. Disclosure outside of Wello.ai is strictly prohibited and is provided to understand
the method of validating claims to reduce employee absenteeism and hospital-acquired infections. Unauthorized sharing or distribution of this document is a
violation of confidentiality agreements and can result in legal action.
A key finding was the enhanced accuracy of ANTE-based
detection in identifying potential cases of illness. This method
proved particularly beneficial in reducing the rate of false
negatives, which is a common challenge in standard fixed
temperature screening methods.
B. Implications of Findings
By focusing on ANTEs, the NWP offers a promising
alternative to traditional health screening methods in the
workplace. It addresses the critical issues of absenteeism,
presenteeism, and HAIs, which are pivotal to maintaining a
healthy workforce and minimizing productivity losses.
Adopting the NWP can lead to more proactive and accurate
health monitoring, enhancing the early detection of potential
infections. This finding is particularly relevant regarding
infectious diseases where identifying symptomatic or
asymptomatic employees early can significantly curb the
infection’s spread within communal settings such as healthcare,
manufacturing, and office facilities.
C. General Claims Supported by Evidence
• The NWP using ANTEs is more effective for the early
detection of potential infectious diseases than CST. The
higher accuracy rate of ANTEs substantiates this claim
in identifying potential cases of illness and significantly
reducing false negatives.
• By tracking ANTEs, NWP provides a personalized
approach to health monitoring. This method effectively
identifies symptomatic and asymptomatic employees
who may be contagious, thus playing a crucial role in
mitigating the spread of infections in workplace
environments.
• The introduction of ANTE-based screening as part of the
NWP aligns with the need for more efficient and non-
disruptive health monitoring methods. This approach
shows promise in reducing absenteeism and
presenteeism, which are key concerns in workplace
health management.
• The findings of this study suggest that a shift towards
more individualized and data-driven health screening
methods, such as NWP, can enhance overall workplace
health and safety while improving productivity.
D. Future Perspectives
The NWP marks a significant step towards advancing health
screening practices. Future research should expand its efficacy
across various demographics and workplaces. Integrating NWP
with other health-monitoring technologies, such as wearable
devices, could lead to an even more advanced approach for the
early detection of health issues. Advancements in AI and
machine learning are expected to refine NWP, offering more
personalized and accurate health assessments. Protocols such as
the NWP in proactive health management will grow, influencing
future occupational and public health strategies.
REFERENCES
[1] Maung Z, Kristensen M, Hoffman B, Jacobson MA. Temperature
screening of healthcare personnel is ineffective in controlling COVID-
19. J Occup Environ Med. 2022;64: 382–384. Available from:
https://www.ingentaconnect.com/content/wk/jom/2022/00000064/00000
005/art00019. doi: 10.1097/JOM.0000000000002518.
[2] Liu Z, Ma Y, Cheng Q, Liu Z. Finding asymptomatic spreaders in a
COVID-19 transmission network by graph attention networks. Viruses.
2022;14. Available from: https://www.mdpi.com/1999-4915/14/8/1659.
doi: 10.3390/v14081659.
[3] Vedaei SS, Fotovvat A, Mohebbian MR, Rahman GME, Wahid KA,
Babyn P, et al. COVID-SAFE: An IoT-based system for automated
health monitoring and surveillance in post-pandemic life. IEEE Access.
2020;8: 188538–188551. Available from:
https://ieeexplore.ieee.org/abstract/document/9220167/. doi:
10.1109/ACCESS.2020.3030194.
[4] Tsai TH, Lu JX, Chou XY, Wang CY. Joint masked face recognition
and temperature measurement system using convolutional neural
networks. Sensors (Basel). 2023. Available from:
https://www.mdpi.com/1424-8220/23/6/2901;23. Internet;23. doi:
10.3390/s23062901.
[5] Zhang L, Zhu Y, Jiang M, Wu Y, Deng K, Ni Q. Body temperature
monitoring for regular COVID-19 prevention based on human daily
activity recognition. Sensors (Basel). 2021;21: 7540. Available from:
https://www.mdpi.com/1424-8220/21/22/7540. doi: 10.3390/s21227540.
[6] Balli S, Shumway KR, Sharan S. Physiology, fever. StatPearls; 2023.
Internet. Available from:
https://www.ncbi.nlm.nih.gov/books/NBK562334/.
[7] Wang Y, Deng Z, Shi D. How effective is a mask in preventing COVID‐
19 infection? Med Devices Sens. 2021;4. Available from:
https://onlinelibrary.wiley.com/doi/abs/10.1002/mds3.10163: e10163.
doi: 10.1002/mds3.10163.
[8] Aronsson G, Gustafsson K, Dallner M. Sick but yet at work. An
empirical study of sickness presenteeism. J Epidemiol Community
Health. 2000;54: 502–509. Available from:
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1731716/. doi:
10.1136/jech.54.7.502.
[9] Szymczak JE. Health care worker presenteeism: A challenge for patient
safety | PSNet. Internet. Available from:
https://psnet.ahrq.gov/perspective/health-care-worker-presenteeism-
challenge-patient-safety.
[10] Webster RK, Liu R, Karimullina K, Hall I, Amlôt R, Rubin GJ. A
systematic review of infectious illness Presenteeism: Prevalence, reasons
and risk factors. BMC Public Health. 2019;19: 799. Available from:
https://bmcpublichealth.biomedcentral.com/articles/10.1186/s12889-
019-7138-x. doi: 10.1186/s12889-019-7138-x.
[11] Hendrich A, Chow MP, Skierczynski BA, Lu Z. A 36-hospital time and
motion study: How do medical-surgical nurses spend their time? Perm J.
2008;12: 25–34. Available from:
https://www.thepermanentejournal.org/doi/pdf/10.7812/tpp/08-
021?download=true. doi: 10.7812/tpp/08-021.
[12] Sowards W. Can the flu be asymptomatic? Internet; 2022.
PassportHealth. Available from:
https://www.passporthealthusa.com/2022/04/can-the-flu-be-
asymptomatic/.
[13] Ryu WS. Virus life cycle. Mol Virol Hum Pathogenic Viruses. 2017;31.
Available from:
https://yonsei.pure.elsevier.com/en/publications/molecular-virology-of-
human-pathogenic-viruses.
[14] Balli S, Shumway KR, Sharan S. Physiology, fever. StatPearls; 2023.
Internet. Available from:
https://www.ncbi.nlm.nih.gov/books/NBK562334/.
[15] Lau LLH, Cowling BJ, Fang VJ, Chan KH, Lau EHY, Lipsitch M, et al.
Viral shedding and clinical illness in naturally acquired influenza virus
infections. J Infect Dis. 2010;201: 1509–1516. Available from:
https://academic.oup.com/jid/article-abstract/201/10/1509/992720. doi:
10.1086/652241.
[16] Facente SN, Hunter LA, Packel LJ, Li Y, Harte A, Nicolette G, et al.
Feasibility and effectiveness of daily temperature screening to detect
COVID-19 in a prospective cohort at a large public university. BMC
Public Health. 2021;21: 1693. Available from:
https://bmcpublichealth.biomedcentral.com/articles/10.1186/s12889-
021-11697-6. doi: 10.1186/s12889-021-11697-6.
This documentation contains confidential information and is not for distribution. Disclosure outside of Wello.ai is strictly prohibited and is provided to understand
the method of validating claims to reduce employee absenteeism and hospital-acquired infections. Unauthorized sharing or distribution of this document is a
violation of confidentiality agreements and can result in legal action.
[17] Mazerolle SM, Ganio MS, Casa DJ, Vingren J, Klau J. Is oral
temperature an accurate measurement of deep body temperature? A
systematic review. J Athl Train. 2011;46. Available from:
https://meridian.allenpress.com/jat/article-abstract/46/5/566/110675:
566–573. doi: 10.4085/1062-6050-46.5.566.
[18] Bai N. Normal body temperature is personal, Stanford Medicine
researchers find | News center | Stanford Medicine. Internet. Available
from: https://med.stanford.edu/news/all-news/2023/09/body-
temperature.html. Stanford Medicine News Center.
[19] McGarry BE, Gandhi AD, Barnett ML. Covid-19 surveillance testing
and resident outcomes in nursing homes. N Engl J Med. 2023;388:
1101–1110. Available from:
https://www.nejm.org/doi/full/10.1056/NEJMoa2210063. doi:
10.1056/NEJMoa2210063.
[20] Zach. How to use Q-Q plots to check normality; 2021. Statology.
Internet. Available from: https://www.statology.org/q-q-plot-normality/.
[21] Shrivastava SR, Shrivastava PS, Ramasamy J. A comprehensive
approach to reduce sickness absenteeism. J Inj Violence Res. 2015;7:
43–44. Available from:
https://pdfs.semanticscholar.org/5601/f49d6af723a6acf6bc9679df7100c
b29525a.pdf. doi: 10.5249/jivr.v7i1.631.