Telehealth, otherwise known as telemedicine, is a method of administering medical care remotely, usually through video conferencing, such as Zoom, or over the telephone. Originally developed to provide basic care to rural and underserved populations, it seen an explosion in use onset by the COVID-19 pandemic (Gajarawala and Pelkowski 2021). Telehealth refers to the use of digital information and video conferencing technologies, such as mobile apps and remote monitoring, to provide healthcare services and information over a distance. It enables patients to access medical consultations, diagnoses, treatments, and follow-up care without the need for in-person visits. Telehealth also supports the ongoing management of chronic conditions and enhances communication between patients and healthcare providers. However, telehealth is not a perfect solution substitute for in-person visits, as certain kinds of care are impossible to administer remotely, suggesting that telehealth is an inferior good relative to in-person visits. In our project, we conducted an analysis of health outcomes associated with telehealth using the 2023 Adult Sample dataset from the NHIS (National Health Interview Survey) where we show that telehealth is not always associated with positive health outcomes. The results from our study show that telehealth is associated with positive health outcomes, but only with respect to specific populations. For others, it is associated with a higher likelihood of being in poor health. For individuals who belong to specific ethnic minorities, such as Hispanic Americans, Alaskan Indians and Native Americans, access to telehealth is associated with a much higher likelihood of being in good health. Additionally, individuals with some chronic diseases, such as hypertension and diabetes, also have a higher likelihood of being in good health with access to telehealth appointments. However, individuals with better access to traditional healthcare, such as those from White and Black populations, have a lower likelihood of being in good health. Similarly, individuals with certain kinds of chronic conditions, such as cancer, have a lower likelihood of being in good health with access to telehealth appointments. Our results suggest that telehealth is an effective form of treatment for some populations, while others would benefit more from traditional in-person medical visits. The implications of our study allow for increased efficiency with respect to health outcomes when considering which kinds of remote appointments physicians should provide. Rather than viewing virtual medical appointments as a substitute for in-person medical appointments, one should only administer them when appropriate.
The healthcare industry has undergone significant changes since the global COVID-19 pandemic. At the onset of the pandemic, many unknowns existed regarding the virus’s nature, the extent of its effects, and its danger to specific populations. Authorities and healthcare providers implemented numerous precautions to protect vulnerable populations, including older adults, individuals with autoimmune diseases, and children. Preventing the spread was of utmost importance; thus, many industries, including the healthcare industry, were forced to adapt. Not only were hospitals, doctor’s offices, and health clinics severely impacted due to the potential for the virus to spread. The issue of keeping healthcare faculty, such as doctors and nurses, from becoming spreaders themselves was just as important as protecting those whom the virus could severely weaken or worse. The solution the healthcare industry found was telehealth visits.
Telehealth offers several benefits that have established new ways to deliver care. One of these benefits is that access to healthcare has expanded. Outside the context of a visit with the doctor, it allows anyone to visit with another person nearly instantaneously, only requiring both parties to have a connection to the internet and a webcam. Without requiring patients to be in person, people who live in rural areas can enjoy access to healthcare professionals. It is important to recognize this convenience and consider it when discussing the importance of telehealth and its implications for those lacking geographical access to care. Another benefit is that telehealth may represent a cost-effective way to deliver patient care by reducing provider overhead. Such visits only require an internet connection and a webcam, rather than the traditional facilities required to run a medical practice. From a patient’s perspective, telehealth represents a more efficient means of receiving care, with zero transportation costs and opportunity costs associated with going to a traditional appointment. Another benefit is the reduced stress associated with visiting the doctor’s office. Telehealth visits could reduce mental stress associated with going to the doctor, helping patients feel more comfortable discussing their symptoms and possibly receiving a higher quality of care. Additionally, receiving news about potential health issues would be easier to process in the comfort of your own home rather than a hospital or medical clinic.
Some shortcomings to consider is that some forms of visits are impossible to hold online. One of the most significant drawbacks of telehealth is that physical examinations are impossible to hold over video conferencing (Ftouni et al. 2022). Such a shortcoming severely limits what kinds of care patients could receive as many symptoms require being in person to diagnose accurately. Another drawback is that some symptoms may go overlooked online. Occasionally, when visiting a medical clinic, one may receive a diagnosis for symptoms they may believe are normal. Such symptoms may not be brought up unless prompted. Another drawback is that telehealth may require education on how to join/participate in virtual appointments. People who require regular consultations with healthcare professionals are often those who have chronic conditions. Among this population there are many elderly who lack education in how to use devices that host video conferencing (Ftouni et al. 2022). Educating those unfamiliar with how to use applications such as Zoom must be considered. An alternative may be providing over-the-phone appointments that do not require such hosting services.
The field of telehealth is not necessarily new; however, its adoption has exploded in recent years due to the COVID-19 pandemic. With nearly three-quarters of physicians working in offices offering these services, it is important to study the health outcomes of individuals who have virtual medical appointments (Henry 2023). Our goal is to evaluate telehealth’s place in the broader healthcare industry, particularly whether telehealth should be considered a substitute for doctor’s visits or a complement to regular checkups for specific chronic conditions.
As discussed by Hersh et al. (2001), the most substantial evidence for individuals who would benefit from telehealth visits are those suffering from chronic conditions, such as people with chronic diseases, hypertension, and AIDS. Additional medical apparatuses could be utilized in the home to specialize their virtual medical appointments to their required care, allowing them consistent care from home. This supports the idea that, on the whole, telehealth should be viewed as a complement to medical appointments that would otherwise have to be regularly taken at a medical clinic, allowing individuals with regular doctor’s appointments the convenience of taking them from home.
As discussed in Dhunnoo et al. (2024), telemedicine is associated with positive health outcomes for people with chronic health conditions, but the sustained effectiveness of this care is unclear. There are challenges associated with telemedicine modality and education regarding how to use video hosting programs, such as Zoom. Such issues should be addressed to enhance telehealth’s accessibility to more populations.
As discussed in Ftouni et al. (2022), telehealth may represent a solution to care disparities, such as with people living in rural areas who typically struggle to find access to traditional health care. By connecting patients with healthcare providers remotely, telehealth alleviates stresses generally associated with finding a doctor traditionally. However, Ftouni et al. (2022) point out that racial and ethnic minorities have less access to telehealth, which should be considered when interpreting our results.
All data for this project came from the 2023 National Health Interview Survey (Moriarity et al. 2022). This survey has been conducted by the CDC National Center for Health Statistics since 1957. In 2023, 29,522 adults were asked a series of around 650 questions for the survey. The questions covered a blanket of health and demographic topics, asking about education, ethnicity, and mental and physical health, among other things. The sample adults were from all over the United States, with over 1,600 counties represented. All responses were gathered into a cross sectional dataset, yielding around 650 variables. The survey dataset is particularly helpful in the context of this research because it includes variables for whether or not somebody went to a virtual doctor’s appointment in the past 12 months, an in person doctor’s appointment in the past 12 months, and it asks for ethnicity. The virtual appointment variable was used to represent telehealth usage in this study. The in person doctor’s appointment variable is useful because it is important to measure the effect of telehealth with and against the established method for obtaining healthcare. Lastly, a main focus of this study is how telehealth may help different demographics, so variables like the one for ethnicity are especially helpful to interact with telehealth in the models.
One issue with the data is that a lot of variables have small sample sizes because they were only asked to adults who responded ‘yes’ to another question. The most egregious example of this is the fact that there are almost 100 variables relating to many different types of cancer that an adult may have and the care they are getting for the disease; but a respondent would only be asked all the follow ups if they responded ‘yes’ to the one question “Do you have cancer?”. For the purposes of this study, we stuck with variables that represented almost the full sample. A list of 59 variables is included in the appendix, these were the ones that we felt represented a good sample and also would have the greatest effect on our dependent variable, good health.
All the variables that we used were either factor or binary. When the dataset variable was a yes or no question, yes was converted to 1, no to 0, and everything else (don’t know, refused, etc.) into N/A. Some variables from the dataset had a range of values, for example good health. The original variable was a scale from 1-5 with 1 representing excellent health, and 5 poor health. We converted this into binary where 1,2,3 would become 1 for good health, and 4 and 5 became 0, representing not good health. Variables like ethnicity and education had a broad scale and we wanted to see the effect of those differences, so we just changed them to factors.
Because our dependent variable is binary and all our independent variables are binary or factor, we used a generalized linear model. The coefficient of a variable X derived from this model can be interpreted as the probability that a respondent will report good health, given X applies to them, ceteris paribus.
As seen in the first appendix regression, we were mainly focused on the effect of virtual doctor’s appointments, in person doctor’s appointments, and ethnicity on good health. The rationale behind choosing these variables over others in the dataset included the number of missing observations and the impact it had on the total number of observations within the model.
The fourth regression from the appendix is our most robust model. We found this model by adding all 59 variables to the basic regression, then sifting out variables that proved irrelevant or did not have enough valid datapoints (marriage stat had over 10,000 missing observations, for example). This left us with a robust regression containing only the most significant variables.
The second model was an attempt to match virtual appointments up against in person appointments. Using the NHIS variables for virtual and in person appointments, four dummy variables were created: only_virtual: (virtual = 1, in person = 0); both: (1, 1); neither: (0, 0); and only_in_person: (0, 1). The variable for if someone only went in person was excluded, so the effect of the other variables is matched up against that one. These were added to the most robust regression because that model was already cleaned up, the original variables for in person and virtual appointments were removed.
The third model includes several interaction terms to look at the relationship virtual appointments have on good health based on things like ethnicity and preexisting chronic conditions. The interaction terms were made by multiplying the NHIS variable for virtual appointments onto other variables representing ethnicity, chronic conditions, etc. This is also an extension of the most robust model for ease of cleaning. The rationale behind including the interaction terms we discussed were the literature, specifically the theorized most positively and negatively impacted populations as outlined in Dhunnoo et al. (2024). Observing the interaction between these populations and access to telehealth visits proved to be some of the most interesting insights our project has corroborated.
All discussion will be focused on models (2), (3), and (4), as model (1) is rudimentary, representing a rough foundation for which our other models to be built upon. Initially, in the most robust model (4), we see that a respondent who reported seeing a doctor virtually in the past year is about 21.7%. less likely to report good health than a respondent who did not have a virtual appointment. Also, model (2) implies that someone who only went to a virtual appointment is about 19% less likely to report good health than someone who visited a doctor in person. Which is supported by the results from model (3), showing the same trend with individuals who have had a virtual medical appointment in the last year being ~22% less likely to be in good health. Another focus of our project was to analyze the effect virtual medical appointments had on chronic conditions. In model (3), we observed that People with hypertension and diabetes who used telehealth are 15.2% and 22.5% more likely respectively to have good health compared to their peers who did not have a virtual appointment. This result in in contrast to individuals who have been diagnosed with cancer being 17.5% more likely to be in poor health, suggesting that telehealth’s effectiveness in contingent on what kind of chronic illness an individual has.
Alternatively, impoverished people using telehealth are 12% more likely to be in good health relative to impoverished people who lack access to telehealth. A final interesting interaction is relating to American Indians and Alaskan Natives. When we interacted virtual appointments with every other available race, the coefficient showed a decrease in good health associated with virtual appointments. However, with the AIAN cohort, those who used telehealth reported good health 86% more than those who did not. It is important to mention here that the excluded ethnicity in the study is the Hispanic population, so all ethnicity variables are in comparison to Hispanic people’s outcomes.
The results of the interaction terms tell us a lot about who benefits from telehealth the most. It seems to be that people who are impoverished benefit a lot from this method, this could be because it is cheaper than seeing a doctor in person. Also, people who have chronic conditions like diabetes and hypertension benefit, possibly because their condition makes it uncomfortable to move around too much or travel, so staying at home on a video call would be better for their health. The last group to talk about in relation to telehealth is the AIAN cohort. This is a group that may live in largely rural areas, and so access to doctors, especially specialists, may not be as easy as it is in a city. Because of this, American Indians and Alaska Natives probably make great use of and could already be well established using telehealth.
It is important to note that all 3 relevant models portray in person doctor’s appointments as beneficial to patient health. In the most robust model, the likelihood that someone reports good health increases by 17.1% if they have had an in person appointment in the past 12 months, and the likelihood actually increases to 18.1% in model (3), which is the only model that shows telehealth having a greater effect than in person. Lastly, model (2) suggests that going in person to the doctors versus virtually increases the likelihood of good health by 19%. The results are unsurprising given how established the healthcare system is, and how in person care has been essentially the only option until the last couple decades.
Some other variables that we found interesting to mention include the education variables. Specifically, people with PhD’s Master’s, and Bachelor’s degrees were 3.3, 3.4, and 2.5 times more likely to report good health than someone who did not complete 12th grade. Also, without the virtual appointment interaction term, the only ethnicity to report worse health than the Hispanic cohort was the Black/African American cohort, implying nationwide these two ethnic groups have the most health problems and probably suffer the most under the current healthcare system. The lowest coefficients in the most robust model were relatively straightforward, relating to having a stroke, hypertension, cancer, and other chronic conditions. One surprising/interesting result was for the difficulty walking variable. The most robust model suggests that people with difficulty walking and/or climbing stairs report good health over 2 times as much as people who are physically comfortable doing these tasks. This finding is difficult to explain, but could have something to do with the lifestyle of someone with difficulty moving being very sedentary.
It is important to note some potential errors in our models. There is a potential issue of heteroskedasticity…. Also, in model 3 it appears that the variable relating to COPD, Emphysema, or Bronchitis Diagnosis is listed in both its binary form and also its unedited form from the original dataset. It is unclear how this effects the model, but we thought it was worth mentioning and could be useful to anyone looking to improve the model.
In conclusion, our study emphasizes the intricate and varied relationship between telehealth and patient health outcomes. While telehealth is a great tool for individuals from specific populations, the same cannot be said for others. Rather than treating it as a uniform treatment for all people, there needs to be consideration for the history of chronic illness and access to traditional healthcare before deciding how much should be offered. Our results have shown that telehealth has particular promise for individuals from rural or underserved communities, such as American Indians and Alaskan Natives, who experience considerably better health outcomes than those without access. Similarly, individuals with chronic illnesses such as hypertension and diabetes experienced better health outcomes, possibly because telehealth better suits their needs for constant and minimally invasive care.
Along with showing the effectiveness of telehealth for specific populations, it is important to recognize the health outcome disparities among ethnic groups. Black/African American and Hispanic populations have reported worse health outcomes across the board compared to other groups. While telehealth may eventually reduce barriers to care, some inequities are not addressed comprehensively and effectively. Addressing such disparities should be a top priority for policymakers and healthcare officials, as an ounce of cure is worth a pound of treatment.
In summary, telehealth represents an excellent opportunity for increased access to efficient care, particularly for populations that align with its strengths. For those who do not, telehealth has a way to go. Ultimately, telehealth should not be considered a substitute for traditional in-person visits, as certain kinds of care are impossible to give virtually. Thus, its application must be carefully considered and integrated alongside in-person visits to ensure equitable and effective treatment across all populations. By tailoring telehealth strategies to patient needs, healthcare systems can optimize their ability to give care and improve health outcomes for all.
Further research should be conducted on a longitudinal scale to measure the long-term effectiveness of telehealth visits. Introducing such a time dimension will also capture changes across populations of interest concerning health outcomes related to telehealth usage and otherwise. Additionally, observing changes in income and telehealth would be an interesting study, as it could capture if the medium behaves like an inferior good when one is below the poverty threshold.
##
## =====================================================================================================
## Dependent variable:
## ------------------------------------------------------------
## Good Health
## (1) (2) (3) (4)
## -----------------------------------------------------------------------------------------------------
## Virtual Appointment in Last Year 0.627*** 0.769*** 0.783***
## (0.585, 0.672) (0.581, 1.016) (0.705, 0.870)
##
## In-Person Appointment 0.498*** 1.171***
## (0.421, 0.589) (0.947, 1.447)
##
## Only Virtual Appointments 0.810***
## (0.487, 1.346)
##
## Both Virtual and In-Person 0.777***
## (0.698, 0.865)
##
## Neither Virtual nor In-Person 0.815***
## (0.648, 1.027)
##
## Virtual*White 0.903***
## (0.678, 1.203)
##
## Virtual*Black 0.826***
## (0.566, 1.206)
##
## Virtual*Asian 0.925***
## (0.555, 1.542)
##
## Virtual*AIAN 1.862***
## (0.516, 6.716)
##
## Virtutal*Other 0.798*
## (0.293, 2.176)
##
## Virtual*Mixed Race 0.464
## (0.169, 1.276)
##
## Virtual*COPD 1.566***
## (1.088, 2.255)
##
## Virtual*Hypertension 1.104***
## (0.896, 1.360)
##
## Virtual*Cancer 0.825***
## (0.634, 1.075)
##
## Virtual*Diabetic 1.232***
## (0.944, 1.607)
##
## Virtual*Impovershed 1.120***
## (0.832, 1.509)
##
## Non Hispanic White 1.209*** 1.388*** 1.449*** 1.390***
## (1.099, 1.330) (1.185, 1.627) (1.214, 1.730) (1.187, 1.629)
##
## Non Hispanic Black/African American 0.792*** 0.951*** 1.026*** 0.953***
## (0.699, 0.898) (0.786, 1.149) (0.830, 1.269) (0.788, 1.152)
##
## Non Hispanic Asian 2.005*** 1.119*** 1.122*** 1.120***
## (1.652, 2.434) (0.863, 1.453) (0.827, 1.522) (0.863, 1.454)
##
## Non Hispanic AIAN 0.729*** 1.076*** 0.869*** 1.075***
## (0.502, 1.059) (0.619, 1.871) (0.517, 1.461) (0.618, 1.869)
##
## Non Hispanic AIAN and Any Other Group 0.530*** 1.186*** 1.208*** 1.189***
## (0.382, 0.734) (0.713, 1.971) (0.637, 2.292) (0.715, 1.977)
##
## Mixed Race 1.741*** 1.211*** 1.709*** 1.213***
## (1.187, 2.555) (0.718, 2.043) (0.836, 3.495) (0.718, 2.050)
##
## High Cholesterol Diagnosis 0.850*** 0.851*** 0.851***
## (0.766, 0.945) (0.768, 0.944) (0.766, 0.945)
##
## Cancer Diagnosisr 0.749*** 0.790*** 0.749***
## (0.658, 0.853) (0.674, 0.926) (0.657, 0.852)
##
## History of Stroke 0.581*** 0.587*** 0.581***
## (0.471, 0.718) (0.476, 0.724) (0.471, 0.718)
##
## Employment Status 1.288*** 1.273*** 1.288***
## (1.157, 1.433) (1.147, 1.411) (1.158, 1.434)
##
## Received Food Stamps in Last Year 0.831*** 0.848*** 0.832***
## (0.717, 0.963) (0.735, 0.978) (0.717, 0.964)
##
## U.S. Born 1.561*** 1.512*** 1.560***
## (1.339, 1.819) (1.303, 1.756) (1.338, 1.818)
##
## Diabetes Diagnosis 0.466*** 0.428*** 0.466***
## (0.410, 0.530) (0.367, 0.499) (0.410, 0.530)
##
## COPD, Emphysema, or Bronchitis Diagnosis 0.495*** 0.413*** 0.495***
## (0.415, 0.589) (0.337, 0.506) (0.415, 0.589)
##
## Received Home Care in Last Year 0.748*** 0.743*** 0.747***
## (0.611, 0.916) (0.608, 0.908) (0.610, 0.914)
##
## Difficulty Walking 2.186*** 2.182*** 2.186***
## (1.956, 2.443) (1.957, 2.433) (1.956, 2.442)
##
## Difficulty Remembering 1.322*** 1.342*** 1.322***
## (1.182, 1.479) (1.203, 1.498) (1.182, 1.479)
##
## Wears Glasses 0.844*** 0.845*** 0.844***
## (0.753, 0.946) (0.757, 0.943) (0.753, 0.945)
##
## Received Prescription in Last Year 0.636*** 0.756*** 0.639***
## (0.538, 0.751) (0.654, 0.874) (0.541, 0.755)
##
## Difficulty Paying Bills 0.648*** 0.660*** 0.648***
## (0.564, 0.746) (0.577, 0.756) (0.563, 0.745)
##
## Uses Hearing Aid 1.308*** 1.309*** 1.307***
## (1.080, 1.585) (1.083, 1.583) (1.079, 1.584)
##
## Had Eye Exam in Last Year 1.195*** 1.210*** 1.196***
## (1.077, 1.327) (1.093, 1.339) (1.077, 1.327)
##
## Received Flu Vaccine in Last Year 1.194*** 1.206*** 1.194***
## (1.080, 1.320) (1.094, 1.330) (1.080, 1.320)
##
## Travelled Outside the U.S. Since 1995 1.213*** 1.213*** 1.213***
## (1.090, 1.350) (1.093, 1.346) (1.090, 1.350)
##
## Issues with Balance or Dizziness 0.738*** 0.735*** 0.738***
## (0.662, 0.822) (0.661, 0.817) (0.662, 0.822)
##
## Pain Some Days 0.770*** 0.763*** 0.770***
## (0.674, 0.881) (0.671, 0.868) (0.674, 0.881)
##
## Pain Most Days 0.487*** 0.472*** 0.487***
## (0.410, 0.579) (0.399, 0.558) (0.410, 0.579)
##
## Pain Every Day 0.362*** 0.360*** 0.362***
## (0.310, 0.423) (0.310, 0.418) (0.310, 0.422)
##
## Below Poverty Threshold 0.646*** 0.631*** 0.647***
## (0.553, 0.755) (0.534, 0.746) (0.553, 0.756)
##
## Straight 0.759*** 0.755*** 0.760***
## (0.530, 1.088) (0.532, 1.071) (0.531, 1.089)
##
## Bisexual 0.540** 0.558** 0.541**
## (0.341, 0.856) (0.357, 0.872) (0.341, 0.857)
##
## Something Else 0.571* 0.606* 0.574*
## (0.301, 1.083) (0.323, 1.138) (0.303, 1.088)
##
## Don't Know 0.535 0.508 0.536
## (0.248, 1.153) (0.247, 1.044) (0.249, 1.154)
##
## Depressed Some Days 0.637*** 0.648*** 0.638***
## (0.561, 0.724) (0.573, 0.734) (0.561, 0.724)
##
## Depressed Half The Time 0.445*** 0.460*** 0.445***
## (0.347, 0.570) (0.361, 0.585) (0.347, 0.571)
##
## Depressed Daily 0.397*** 0.409*** 0.398***
## (0.303, 0.521) (0.315, 0.530) (0.303, 0.522)
##
## Biological Sex 0.713*** 0.723*** 0.713***
## (0.647, 0.786) (0.658, 0.794) (0.647, 0.786)
##
## No Diploma 1.184*** 1.118*** 1.184***
## (0.856, 1.638) (0.820, 1.525) (0.856, 1.639)
##
## GED or Equiv 1.311*** 1.290*** 1.312***
## (0.968, 1.776) (0.964, 1.726) (0.969, 1.777)
##
## High School Grad 1.560*** 1.533*** 1.562***
## (1.306, 1.864) (1.293, 1.818) (1.307, 1.866)
##
## Some College 1.754*** 1.724*** 1.755***
## (1.443, 2.131) (1.429, 2.080) (1.445, 2.133)
##
## Occupational Assoc 1.635*** 1.546*** 1.636***
## (1.251, 2.138) (1.196, 1.998) (1.251, 2.139)
##
## Associate Degree 1.751*** 1.766*** 1.752***
## (1.403, 2.186) (1.426, 2.188) (1.404, 2.187)
##
## Bachelor's 2.525*** 2.526*** 2.530***
## (2.065, 3.088) (2.081, 3.066) (2.069, 3.093)
##
## Master's 3.430*** 3.476*** 3.430***
## (2.680, 4.390) (2.733, 4.421) (2.681, 4.390)
##
## PhD 3.346*** 3.461*** 3.347***
## (2.354, 4.755) (2.447, 4.896) (2.355, 4.758)
##
## Internet Access at Home 1.279*** 1.278*** 1.277***
## (1.095, 1.494) (1.099, 1.486) (1.094, 1.492)
##
## Hospitalized Overnight 0.585*** 0.584*** 0.585***
## (0.510, 0.671) (0.510, 0.669) (0.510, 0.671)
##
## Hypertension Diagnosis 0.645*** 0.628*** 0.645***
## (0.578, 0.719) (0.553, 0.713) (0.579, 0.720)
##
## Constant 10.017*** 5.869*** 5.033*** 4.960***
## (8.386, 11.966) (3.636, 9.473) (3.174, 7.980) (2.988, 8.234)
##
## -----------------------------------------------------------------------------------------------------
## Observations 25,373 22,755 24,519 22,755
## Log Likelihood -11,172.100 -6,245.322 -6,669.443 -6,245.727
## Akaike Inf. Crit. 22,362.200 12,596.650 13,462.890 12,595.450
## =====================================================================================================
## Note: *p<0.1; **p<0.05; ***p<0.01Stevens and Donohue-Ryan (2021)