Dog Owners' Survey reveals Medical Alert Dogs can alert to multiple conditions and multiple people.

Medical Alert Dogs (MADs) are a promising support system for a variety of medical conditions. Emerging anecdotal reports suggest that dogs may alert to additional health conditions and different people other than those that they were trained for or initially began alerting. As the use of medical alert dogs increases, it is imperative that such claims are documented empirically. The overall aims of this study were to record the proportion of MAD owners who have a dog that alerts to multiple health conditions or to people other than the target person and to determine whether any sociodemographic variables were associated with dogs alerting to multiple conditions, multiple people, or both. MAD owners completed an online survey that contained a series of forced choice questions. Sixty-one participants reported a total of 33 different conditions to which dogs alerted. Eighty-four percent of participants reported that their dog alerted to multiple conditions and 54% reported that their dog alerted to multiple people. This is the first study to document that a large percentage of people report that their MAD alerts to multiple conditions and/or to multiple people. We present a discussion of how these alerting abilities could develop, but questions about the underlying mechanisms remain.

Introduction

Over the past several decades, mounting evidence has emerged for the health and well-being benefits associated with dog ownership [1-3] although this has not been universally accepted [4]. Benefits of dog ownership may manifest in several ways, including decreasing loneliness [5], providing emotional support through physical or psychologically difficult times (e.g., [6-8]), increasing the number of social contacts [9] and increasing activity levels as well as overall health (e.g. [10]). One way in which dogs have the potential to substantially impact their owners’ health is by alerting their owner to physiological changes. This is when a dog detects changes in the owner’s physiology and the owner is made aware of this through changes in their dog’s behaviour. Dog owners have reported behaviour changes in their dogs prior to migraines [11], decreases in blood sugar levels [12,13], the onset of seizures [14,15], and behaviours directed towards areas affected with cancer ([16] c.f. [17]). In these instances, owners also report changes in their dogs’ behaviour even before they recognise their own associated symptoms. For example, Williams and Pembroke [16] were the first to report on a woman whose dog showed persistent interest in a skin lesion that later turned out to be malignant melanoma, and in a study conducted by Wells et al. [13], 33.6% of dog owners with diabetes reported changes in their dogs’ behaviour before they themselves recognised their own symptoms associated with hypoglycaemia. Taken together, these findings suggest that dogs are detecting something not yet perceptible to the person [13,18].

Given dogs’ olfactory acuity [19-21], it is postulated that dogs are detecting volatile organic compounds (VOCs) associated with changes in their owners’ physiology [22]. VOCs are chemical compounds that have a high vapour pressure and thus exist in gaseous form at room temperature. Within the human body, endogenous VOCs are produced during metabolic processes of cells, are released through breath and are present in the headspace of sweat, urine, faeces, and blood [23]. Analytical chemistry has revealed that specific patterns of VOCs have the potential to serve as odour biomarkers for metabolic conditions and disease states [24], including, but not limited to, cancers [25], diabetic hyper- [24] and hypoglycaemia [26], asthma [27], and epileptic seizures [28]. Empirical studies with olfactory detection dogs support these findings; dogs have been shown to detect odours associated with many of the same conditions including, cancers [29], epileptic seizures [30], and hypo- and hyperglycaemia [31,32].

Following such reports, charities and training organizations have begun to harness dogs’ ability to detect odour cues associated with physiological changes in humans. Medical Alert Dogs (MADs) are now trained and placed in homes to alert people to a range of health conditions, including diabetes (hypo- and hyperglycaemic episodes; [12,33,34], epileptic seizures [35], asthma attacks [36], allergic reactions [37,38], Addison’s disease [39,40] and Postural Orthostatic Tachycardia Syndrome (POTS) episodes [41]. Specific training protocols vary across charities, training establishments and owners who train their dogs themselves. Despite this, the basic protocol is usually a version of teaching the dog their ‘target odour’ (e.g., breath or sweat samples collected when that person experiences the medical condition that the dog is intended to alert to) followed by the shaping of ‘alerting’ behaviours appropriate to communicate to the owner the presence of the target odour. It is important to note, however, that dogs may employ a range of behaviours to communicate an alert, even if they have undergone the same training protocols [34]. Some owners may prefer specific ‘alert’ behaviours such as picking up a bringsel (a short item suspended from the collar of a detection dog that the dog takes in its mouth as an alert behaviour) or fetching a medical bag, whereas others may utilise behaviours such as the dog staring, nuzzling, or pawing. Some MADs receive no formal training and the dog seemingly spontaneously begins to ‘alert’ during, or immediately prior to, episodes of their owner’s medical condition (e.g., [34]). In this instance, owners may establish these behaviours, through reinforcement, over repeated exposures. It is possible that certain aspects of the dogs’ training, for example where they were trained, or how their alerts are responded to, could predict certain behavioural outcomes.

Despite the increased popularity of MADs, a detailed understanding of exactly what a dog is responding to, and the full impact of the owner-dog training interaction, is not yet fully understood. A noteworthy phenomenon is emerging whereby MAD owners report that their dog alerts to health conditions outside of what they were originally trained to alert, or first began alerting. For example, many dogs trained to alert to hypoglycaemia begin spontaneously alerting their owner to hyperglycaemia [34,42]. Furthermore, dogs trained to alert to hypoglycaemia have been reported to alert both their owner and/or other people to other health conditions such as anxiety or asthma attacks (Olivia Rockaway, February 2020). This phenomenon has yet to be documented empirically and is thus the focus of the current study.

Therefore, using an online survey we conducted an exploratory study on a sample of MAD owners. The aims of this study were to document sociodemographic information for MADs and their owners, to document the proportion of MAD owners that report that their dog alerts to multiple conditions and/or multiple people, and to determine whether any sociodemographic variables were associated with whether or not a dog alerted to multiple conditions, multiple people, or both.

Materials and methods

Study design

This study utilised a cross-sectional, retrospective design. Participants completed an online survey (described below) that was prepared and presented to participants through the online survey platform, Qualtrics.

Participants and recruitment

Recruitment

Participants were recruited worldwide through advertisements on social media websites (e.g., Facebook, Instagram, and Twitter) related to dog ownership, service dogs, and relevant medical conditions. Responses were collected from May 2019 until August 2019.

Inclusion criteria

Prospective participants for this study were required to be at least 16 years of age, be able to complete the survey in English, have access to a device to complete the study online, and have a dog that alerted themselves, or someone they cared for, to a medical condition. The dog could be trained specifically for medical alert or could have started alerting without any formal training.

Survey

A bespoke survey was designed to assess owner and dog sociodemographic variables, as well as the conditions and people to whom the dogs alerted (available upon request to the corresponding author). The survey consisted of three sections of questions. In the first section, participants were asked about the nature of their relationship with the alerting dog (do they: own a trained medical alert dog that alerts to themselves, own a dog that alerts to themselves without formal training, care for a person that has a trained medical alert dog, or care for a person who has a dog that alerts a medical condition without any formal training), and if the dog was formally trained, who trained the dog. Participants were also asked if their dog receives maintenance training to continue the alerting behaviour, and if so, who completes the maintenance training (a charity or organisation trainer, a private trainer, themselves). These questions were followed by sociodemographic questions including the gender and age of the person to whom the dog alerted, the age, sex, and breed of the dog, how long the dog and the target person had lived together, the relationship between the target person and the dog (the target person does not like their dog, the target person is indifferent towards their dog, the target person sees the dog as a pet and good companion, the target person and dog are best friends), and how friendly the dog is towards people other than the target person (the dog does not like other people, is indifferent towards other people, likes attention from other people, or loves attention from other people). Participants whose dogs began alerting without any formal training were asked how soon after getting their dog did it begin alerting to a medical condition (less than six months, six months to a year, one to three years, three to five years, or greater than five years). Participants who were completing the survey for a dog formally trained for medical alert were asked whether the dog was trained for any other specialised professions (e.g., visual guide dog, hearing dog, police, search and rescue).

The second section of the survey consisted of questions about the condition(s) to which the dog alerted, and how accurate the participant felt the dog was at alerting to the conditions. In this section, participants were asked to identify the first condition to which the dog was trained to alert, or first began alerting. Participants were presented with a matrix question that consisted of a list of conditions (seizures, hypoglycaemia, hyperglycaemia, anxiety, allergic reactions, narcolepsy, migraines, asthma, Addison’s disease, and POTS), as well as an option to enter a different condition (if not listed), alongside categories of ratings (0–25%, 25–49%, 50–74%, 75–100%) for participants to identify how accurate they felt the dog alerted to a particular condition.

The third section of the survey asked participants whether the dog alerted people other than the person for whom the dog was trained to alert (or first began alerting to), and to what condition(s) the dog alerted other people. Participants that indicated that their dog alerted to people other than the target person were presented with a matrix question that consisted of a list of people [family member(s) in the same home, family member(s) in a different home, friend(s), stranger(s)] as well as ‘other’ with the option of entering text, alongside a number rating of how many of each kind of person the dog alerts (one, two, three, and four or more). In a subsequent question, participants were asked, to the best of their knowledge for what condition(s) their dog had alerted these other people (presented as a list the same as the above-mentioned list of conditions, including an option to add different conditions).

Procedure

Participants completed the survey on a personal electronic device that had access to the internet. Prospective participants followed a link to the survey page where they were first presented with the participant information sheet that provided information regarding the aims of the study, what kinds of questions participants would be asked to answer, and how their data would be used and stored. Participants were then presented with the informed consent form. Participants were required to provide consent by indicating that they agreed to a series of statements prior to starting the survey. Participants were advised that they were welcome to withdraw their participation at any time by closing their internet browser.

Ethical approval

The study protocol was approved by the Faculty Research Ethics Committee at Queen’s University Belfast (EPS 19_98), and the Research Ethics Board at Dalhousie University, Canada (REB 2019–4803).

Data analyses

Chi-square tests were used to determine whether any significant associations existed between the variables listed in Table 1 and whether or not a dog alerted to multiple conditions, multiple people, and both multiple conditions and multiple people. Fisher’s exact tests were used for 2x2 tables due to the small cell sizes, and for data that violated the assumption that the cell is expected to be a value of 5 or more in at least 80% of the cells, and/or that no cell had an expected value of less than one [43]. Due to the high number of comparisons, Šidák corrections were used to avoid false positives in each family-wise set of comparisons. All data were analysed using IBM SPSS Statistics (Version 24).

Table 1

Demographic variables of Medical Alert Dogs and the target person to whom the dog alerts.

All Dogs	n	Freq	%
Nature of relationship with dog	64
Own a trained MAD that alerts to self		31	49%
Care for a person that is partnered with a trained MAD		2	3%
Own a dog that alerts to self without formal training		27	42%
Care for a person that is partnered with a dog that alerts without formal training		4	6%
Is the dog formally trained for medical alert	64
Yes		33	52%
No		31	48%
Who trained the dog	33
Charity		8	24%
Private Trainer		4	12%
Self		21	64%
Does the dog receive maintenance training	60
Yes		31	52%
No		29	48%
Who does the maintenance training	31
A charity/organisation trainer		3	10%
A private trainer		4	13%
Self		24	77%
Gender of target person to whom dog alerts	64
Male		10	16%
Female		54	84%
Age of target person to whom dog alerts	64
Child (5–14)		2	3%
Youth (15–25)		14	22%
Adult (25–64)		43	67%
Senior (65+)		5	8%
Sex of dog	64
Male		33	52%
Female		31	48%
Dog purebred or mixed breed	64
Purebred		49	77%
Mixed breed		15	23%
How long the target person has been paired with the dog	64
Under 1 year		9	14%
1–3 years		29	45%
4–6 years		11	17%
7–10 years		10	15%
>10 years		5	8%
Target person’s feelings towards dog	64
Does not like the dog		0	0
Indifferent towards the dog		0	0
The dog is a pet and good companion		12	19%
The target person and dog are best friends		52	81%
Friendliness of dog towards people other than target person	64
Does not like other people		1	2%
Indifferent towards other people		9	14%
Likes attention from other people		24	37%
Loves attention from other people		30	47%
Dogs with No Formal Training for Medical Alert	n	Freq	%
How much time with dog before dog began alerting	31
<6 months		13	42%
6 months—1 year		8	26%
1–3 years		3	10%
3–5 years		5	16%
>5 years		2	6%
Dogs Formally Trained for Medical Alert
Is the dog trained for other specialised activities	33
Yes		18	54%
No		15	45%

Results

Demographics

A total of 72 people consented to participating in the survey, however eight participants withdrew from the study providing no usable data. A further three participants provided demographic data but did not report on the conditions to which or people to whom the dog alerted. Full data sets were provided by 61 participants. A large majority of participants were from North America (72%), followed by the United Kingdom (20%), Europe (6%), and Australia (2%).

Most participants reported on dogs that alerted to themselves (91%). Just over half (52%) of the dogs reported on were trained for medical alert while 48% of dogs started alerting to medical conditions without any prior training. A majority (64%) of dogs trained for medical alert were trained by the owner. Just over half of participants (52%) reported that their dog received training to encourage it to continue alerting to physiological changes, with the majority of the training being conducted by the owners themselves (77%). Most participants (84%) reported on a dog that alerted to a target female person and the majority of respondents (67%) reported that the target person was between the ages of 24 and 64.

Participants reported on 52% male dogs (48% female dogs), 77% of dogs were purebred, and the average age of the dogs was 63 months (SD = ±41, minimum age = 7 months, maximum age = 172 months). Just under half (45%) of participants reported that the dog had been paired with its target person for between one and three years. A large proportion (81%) of participants reported that the dog and the target person were best friends, and the large majority of participants reported that their dog liked (37%) or loved (47%) attention from people. Of those dogs that were trained for medical alert, 54% were also trained for other specialised activities (e.g., guide dog, psychiatric assistance/emotional support, search and rescue). For those dogs that began alerting without any formal training for medical alert (n = 31), 42% of participants reported that the dog began alerting within the first 6 months of having the dog. The total distributions of these demographics can be seen in Table 1.

Conditions to which dogs alert

Participants reported a total of 33 different conditions to which dogs alerted (listed in Table 2). The most common conditions that dogs were reported to alert to were anxiety, hypoglycaemia, hyperglycaemia, migraines, seizures, and POTS. A large majority (84%) of participants reported that their dog alerted to more than one condition (Table 3), with the average number of conditions that dogs were reported to alert to being M = 2 (SD = 1.66, Min = 1, Max = 9). Dogs that alerted to anxiety, hypoglycaemia, hyperglycaemia, migraines, seizures, or POTS were also frequently reported as alerting to other conditions, however, Fishers exact tests with a Šidák correction for multiple comparisons (original α = .05 αSID = .009) revealed that alerting to any of these six conditions was not significantly associated with whether or not a dog alerted to multiple conditions (anxiety p = .092, hypoglycaemia p = .489, hyperglycaemia p = .144, migraine p = .485, seizure p ≈ 1.000, and POTS p = .184; for those dogs that alerted to anxiety, hypoglycaemia, hyperglycaemia, migraines, seizures, or POTS, the other conditions that they were reported to alert to are listed in the supplementary materials). When asked to report on the accuracy with which dogs alerted to conditions, owners perceived their dogs to be highly accurate, with over 70% of ratings being between 75%-100% (see Table 4).

Table 2

Conditions to which owners report their dogs alert.

Condition	Number of times condition was reported	People that reported this condition that also reported that the dog alerts to multiple conditions		People that reported this condition that also reported that the dog alerts to multiple people
		n	%	n	%
Anxiety	28	26	93%	16	57%
Hypoglycaemia	27	24	89%	18	67%
Hyperglycaemia	20	19	95%	12	60%
Migraine	19	17	89%	12	63%
Seizure	11	9	82%	4	36%
POTS*	11	11	100%	5	45%
Allergic reaction	4	4	100%	3	75%
Narcolepsy	4	4	100%	2	50%
Asthma	3	3	100%	3	100%
Periodic paralysis	3	3	100%	3	100%
Arthritis	3	3	100%	2	67%
Cancer	2	2	100%	2	100%
Dissociative episodes	2	2	100%	2	100%
Dystonia	2	2	100%	1	50%
Heart complications	2	2	100%	2	100%
Muscle spasms	2	2	100%	2	100%
Addison’s Disease	1	1	100%	1	100%
Ankle sprain	1	1	100%	1	100%
Blackout	1	1	100%	1	100%
Cataplexy	1	1	100%	1	100%
Ehlers-Danlos Syndrome	1	1	100%	1	100%
Pancreatitis	1	1	100%	1	100%
High blood pressure	1	1	100%	1	100%
Knee injury	1	1	100%	1	100%
Low oxygen	1	1	100%	1	100%
Sepsis	1	1	100%	1	100%
Sinus tachycardia	1	1	100%	1	100%
Depression	1	1	100%	0	0
Cluster headache	1	1	100%	0	0
Postural hypotension	1	1	100%	0	0
PTSD†	1	1	100%	0	0
Sleep apnea	1	1	100%	0	0
Syncope	1	1	100%	0	0

*POTS: Postural Orthostatic Tachycardia Syndrome

† Post Traumatic Stress Disorder.

Table 3

The percentage of participants that report that their dog alerts to multiple conditions, multiple people, and both.

		Dog Alerts to Multiple People n (%)
		Yes	No	Total
Dog Alerts to Multiple Conditions %(n)	Yes	28 (46%)	23 (38%)	51 (84%)
	No	5 (8%)	5 (8%)	10 (16%)
Total		33 (54%)	28 (46%)	61 (100%)

Table 4

Accuracy with which dogs are reported to alert to condition(s).

Accuracy Rating	Dog alerts to multiple conditions n (%)	Dog alerts to a single condition n (%)
0–24%	8 (6.2%)	0
25–49%	13 (10%)	0
50–74%	11 (8.5%)	1 (10%)
75–100%	98 (75%)	9 (90%)

Fisher’s Exact tests were completed for all variables in Table 1 (except the nature of the relationship with the alerting dog) and whether or not a dog alerted to multiple conditions. Analyses revealed that, when a Šidák correction for multiple comparisons was applied (original α = .05, αSID = .004), for dogs without formal training for medical alert, the amount of time the primary person had been with their dog before it began alerting was marginally significantly related to whether or not the dog alerted to multiple conditions (see supplementary materials). The highest proportion of dogs that alerted to multiple conditions alerted within the first 6 months, 42% (13/31), followed by 26% (8/31) between 6 months to a year, 16% (5/31) 3 to 5 years, 10% (3/10) 1 to 3 years, and 6% (2/31) greater than 5 years (see Table 5) (p = .004, two-sided Fisher’s Exact Test). Whether or not the dog was formally trained for medical alert, who trained the dog, whether or not the dog received maintenance training, and who conducted the maintenance training were not significantly associated with whether or not the dog alerted to multiple conditions. Furthermore, the age of the target person to whom the dog alerted, the sex of the dog, how long the target person had been with their dog, the target person’s feelings towards the dog, the friendliness of the dog towards people other than the target person, were not significantly associated with whether or not a dog alerted to multiple conditions. For those dogs that were formally trained for medical alert, whether or not the dog was trained for other specialised activities were not significantly associated with whether or not the dog alerted to multiple conditions (see supplementary materials).

Table 5

The amount of time a dog with no formal training for medical alert spent with a target person before the dog began alerting to medical conditions, and whether or not the dog alerts to multiple conditions.

		Amount of time target person spent with dog before dog began alerting n(%)
		<6 months	6 months -1 year	1–3 years	3–5 years	>5 years
Does the Dog Alert to Multiple Conditions?	Yes	12 (39%)	8 (26%)	1 (3%)	4 (13%)	0
	No	1 (3%)	0	2 (6%)	1 (3%)	2 (6%)

People to whom dogs alert

Just over half of participants (54%) reported that their dog alerted medical conditions to people other than the target person (see Table 3). Dogs were reported to alert to family members in the same home, family members in different homes, friends, and strangers (see Table 6). These were instances of the dog alerting someone (other than their owner) to a health condition specific to that person, rather than the dog alerting another person to get attention for their target person. Considering the top six conditions that dogs were reported to alert to, Fishers exact tests with a Šidák correction for multiple comparisons (original α = .05, αSID = .009) revealed that there were no significant associations between whether or not a dog alerted to these conditions and whether or not a dog alerted to multiple people (anxiety, p = .797, hypoglycaemia, p = .121, hyperglycaemia, p = .591, migraine, p = .412, seizure, p = .317, and POTS, p = .740). The conditions to which dogs were reported to alert to other people and the number of participants that reported their dog alerted other people to those conditions can be seen in Table 7. Of the 33 respondents that reported that their dog alerted to multiple people, only 6% (n = 2) reported that their dog alerted other people to condition(s) other than conditions to which the dog alerted the target individual (see supplementary materials), while 94% of participants reported the dog alerted other people to the same conditions to which it alerts the target individual.

Table 6

The number of participants that reported that their dog alerted to 1, 2, 3, or 4 or more people other than the target person.

Other people to whom dog alerts	1	2	3	4 or more
	No. of participants
Family member in the same home	11	4	1	2
Family member in a different home	4	3	0	4
Friends	5	5	4	5
Strangers	4	6	0	9

Table 7

The conditions to which dogs were reported to alert to other people and the number of participants that reported their dog alerted other people to those conditions.

Condition	No. of participants
Hypoglycaemia	15
Anxiety	12
Hyperglycaemia	6
Migraine	5
Cancer	2
Seizure	2
Dissociative episodes	1
Heart attack	1
POTS*	1
Allergic reactions	1
Low Oxygen	1
Dystonic spasms	1
Knee injury	1
Ankle sprain	1
Shoulder spasm	1
Periodic paralysis	1
Asthma	1
High blood pressure	0

*POTS: Postural Orthostatic Tachycardia Syndrome.

Fisher’s Exact Tests revealed no associations between sociodemographic variables and whether or not a dog alerted to multiple people (see supplementary materials).

Dogs alerting to both other conditions and other people

Nearly half (46%) of participants reported that their dog alerts to both other conditions and other people (Table 3). For each condition reported, the percentage of people that reported that their dog alerted to that condition and also reported that the dog alerted to multiple conditions or multiple people can be seen in Table 2. Considering the top six conditions that dogs were reported to alert to, Fisher’s Exact Tests with a Šidák correction for multiple comparisons revealed no significant associations between these conditions and whether or not a dog alerted to both multiple people and multiple conditions (anxiety, p = .612, hypoglycaemia, p = .075, hyperglycaemia, p = .172, migraine, p = .270, seizure, p = .526, and POTS, p = .1.000). Furthermore, additional Fisher’s Exact Tests with a Šidák correction for multiple comparisons revealed no associations between the sociodemographic variables and whether or not a dog alerts to both other conditions and to people other than the target person (see supplementary materials).

Discussion

The aims of this study were to document sociodemographic information for MADs and their owners, to document the proportion of MAD owners that report that their dog alerts to multiple conditions and/or multiple people, and to determine whether any sociodemographic variables were associated with whether or not a dog alerted to multiple conditions, multiple people, or both. Participants completed an online survey that gathered sociodemographic information for the target person and the dog, and had participants report on the conditions to which and people to whom the dog alerts. The main findings were that participants reported a total of 33 different conditions to which their dogs alerted, a large majority of dog owners reported that their dog alerts to multiple conditions (84%), and over half of respondents reported that their dog alerts to multiple people (54%). Just under half of participants (46%) reported that their dog alerts to both other conditions and other people. This is the first study of its kind to document the phenomenon of dogs, both specially trained for medical alert, and dogs with no specific training for medical alert, alerting to multiple different conditions, and to multiple different people.

Although the aim of this study was not to determine how, or why, dogs may alert their owner, and others, to multiple health conditions, we will discuss possible factors that may contribute to this phenomenon and how these factors could explain our findings. First, it could be the case that a similar physiological state precedes a number of the conditions reported and that the dogs are detecting this preceding state. For example, the most commonly reported condition in this study was anxiety and of those who reported that their dog alerted to anxiety, 93% of participants reported that their dog also alerted to other conditions (however this relationship was not significant). Furthermore, although there was no significant association between whether or not a dog alerted to anxiety and whether or not a dog alerted to multiple people, anxiety was the second most frequently reported condition that dogs alerted to other people. There is a known association between anxiety and epilepsy [44], and there is evidence for a link between anxiety and migraines [45,46]. It is possible that an individual experiences anxiety or stress before a seizure, migraine, or other physiological changes and their dog simply detects the stress preceding any host of physiological changes. Participants also reported that their dogs have alerted to injuries such as sprains. Although a dog can only be aware of an injury after it happens, the physiological and emotional stress experienced by the person along with the injury could signal to the dog that their owner is experiencing a change in overall wellbeing. While there is some evidence that psychological stress produces detectable VOCs [47] and that dogs can detect odour cues associated with fear [48] and stress [49], the claim that stress precedes multiple conditions requires further investigation.

Next, although analytical chemistry has revealed potential VOC profiles for a wide number of conditions, different analyses of the same condition often yield differing results [50,51] and conversely, analysis of different conditions reveal overlap in VOCs [52]. Analysis of VOCs is complicated by the fact that there is large inter-person variability in VOCs; differences in diet, medications, metabolism, and environmental exposures can result in variability in endogenous and exogenous VOCs both within and between people [53]. Overall, defining the VOC profiles for individual health conditions is still ongoing and there is need for further understanding of how these profiles may be similar or different between individuals and how this may translate to the behaviour of a dog. It is possible that the same, or closely related, VOCs are emitted across different conditions or by different people, and dogs that alert to multiple conditions and/or multiple people are detecting these similarities. In this case, a dog may provide an alerting response because the odour is ‘close enough’ to the target odour. Whether or not a dog considers an odour to be ‘close enough’ is the second factor that could explain a dog alerting to multiple conditions or multiple people.

MADs are exposed to a continuous stream of odours at all times. Therefore, the detectability of the VOCs associated with specific physiological changes can be obscured by the background ‘noise’ of other odours. One component of a dog’s ability to detect a specific odour is their biological ‘hardware’. Dogs have upwards of 200 million olfactory receptors to which odorants bind and where signal transduction to the brain occurs [54]. However, the exact number of olfactory receptor cells in a dog’s nasal cavity is variable and dependent upon dog breed and genetics [55]. Furthermore, the genes that code for these receptors have documented allelic variation [55] and studies suggest that a dog’s ability to detect a target odour is related to particular alleles [54]. As such, a dog’s ability to detect specific odour(s) associated with a condition and therefore alert to the condition may be, in part, due to their underlying biology.

Combined with their perceptual ability, other factors that could impact whether or not a dog alerts to a medical condition are affective and temperamental differences in emotional states impacting attention and decision making [56-59]. In other words, each dog will have a threshold for whether they perform the alert behaviour to a target odour, or not, when they are uncertain about the choice to make. Signal Detection Theory [60] terms this threshold an individual’s ‘criterion’ and each individual’s criterion exists on a continuum from conservative to liberal. Considering MADs, a more conservative dog could be one that only alerts when they are confident that an odour represents the condition that they have been trained to alert to or have been reinforced for alerting to previously. Conservative dogs might be less likely to alert when the condition is not being presented (minimising false alarms) and could be more likely to withhold an alert when the condition is in fact present (thereby committing more misses). Following this logic, conservative dogs may be less likely to alert to multiple conditions or multiple people, as they would be less likely to respond to odours that are similar, but not exact to, those on which they were trained. On the other hand, a liberal MAD would be more likely to alert to odours that are ‘close enough’; odours that only approximate the condition of interest. As such, liberal dogs could be more likely to commit false alarms but less likely to miss instances of a condition. In the context of the current study, a liberal MAD would be more likely to generalise their alert response to odours that are not exact to those to which it had previously been reinforced, therefore being more likely to alert to multiple conditions and/or multiple people. Empirical tests in controlled laboratory settings have allowed researchers to measure the parameters of Signal Detection Theory and have highlighted individual differences in dogs’ sensory perceptual abilities and decision criteria during human odour detection tasks [32,61]. Although a dog may have a natural inclination to be a conservative or liberal decision maker, their decision-making bias can be manipulated through training and reinforcement. Therefore, an additional factor that may influence the likelihood that a dog alerts to multiple conditions, or multiple people, is their reinforcement history.

A dog’s reinforcement history could play a large role in the likelihood of them demonstrating alerting behaviours to multiple conditions or people. Most MADs are trained using positive reinforcement for a correct alert (e.g., they receive praise or a treat after successfully alerting their owner to a condition). By definition, reinforcing the alert behaviour will increase the likelihood that the behaviour will occur again in the future [62]. A liberal dog may present the alerting behaviour in different contexts and to different odours. If the alert behaviour occurs when the owner or another person is experiencing a physiological change and if the alert behaviour is then reinforced properly, the dog can become conditioned to respond to a different health condition or to a different person. Similarly, if an owner demonstrates particular behaviours associated with the onset of a condition, and these behaviours elicit attention seeking or affiliative behaviours from their dog, the owner may, over time, interpret the dog’s behaviours as an alert and reward them accordingly. As such, the owner has again conditioned an alerting response to a new condition. This same process can be applied to dogs being rewarded for trying their alert behaviour directed to another person.

It should be noted that fifty-two percent of owners in the study sample reported that their dog receives ongoing maintenance training to continue alerting. It is important to consider what may constitute ‘training’. An owner may have interpreted this term as meaning formal engagement with specific training methods, or hiring a trainer, for example. However, it can be argued that any form of reinforcement following a correct alert is a form of training, in that you are increasing the likelihood of that behaviour occurring in future. Despite only fifty-two percent of owners reporting that their dog receives ongoing training, the vast majority of owners reported that their dog was highly accurate at alerting (75–100% correct). Wilson et al. [34] found that Diabetes Alert Dog owners who were most compliant with their ongoing training protocol (e.g., correctly rewarding true positives and ignoring false positives) had better performing dogs than those owners who did not comply with protocol. It is possible that there was a misinterpretation of the word ‘training’ in this instance, and that more owners are positively reinforcing their dog’s alerts than were reported. Alternatively, or additionally, it is possible that our obtained reports of accuracy are biased by the fact that it is an owner reporting on their own dog. This phenomenon has been established previously within Diabetes Alert Dog owners, where owner reports of accuracy were higher than objective measures of performance (e.g. [33]) The current study includes MADs who, all together, are reported to alert to 33 different health conditions, with many dogs reported to have received no formal training, receive no ongoing training, and yet provide high levels of alerting accuracy. Results such as these highlight the need for objective assessments of MAD performance and behaviour in future studies. However, as this was a preliminary and exploratory study, the possible bias in owner-reporting was accepted as an aspect of documenting this phenomenon.

Analyses revealed that the only marginally significant association was that, for dogs with no previous formal training for medical alert, the amount of time the dog had spent with the target person before it began alerting was significantly related to whether or not the dog alerted to multiple conditions. The distribution of data here would suggest that dogs that began alerting their target person within six months of being together were more likely to alert to multiple conditions than those dogs that spent over six months with their target person before they began alerting. As these dogs received no formal training, they initiated ‘alert’ behaviours which were likely then developed into consistent alerts through the owner’s response to the alerting behaviour over time. Given that this was not a trained response, these dogs likely had a strong natural response to either the owner’s odour profile, or behavioural changes associated with a health-related episode, which then resulted in changes in their own behaviour. The dogs’ behavioural responses to their owners’ physiological changes may have been differentially reinforced for some conditions and not others. There are any number of reasons why this could have occurred, including whether or not the owner actually linked the dogs’ behaviour to a specific physiological change, or whether the owner did not care about or need their dog alerting to a specific physiological change and therefore did not reinforce the dogs’ behaviour. It could be assumed that this process would take place at the beginning of the human-dog relationship and, once the reinforcement contingencies were in place, the dog learned to discriminate between different conditions and only alert some conditions and not others. Additionally, it is possible that as a dog spends more time with a particular owner, they develop a more thorough understanding of what the owner smells like and what specific odours signify a significantly altered physiological state in that owner.

Analyses revealed that none of the variables assessed were related to whether or not a dog alerted to multiple people. Of interest, however, was that of the thirty-three dogs who were reported to alert to other people, the vast majority of participants reported that dogs alerted the other people to the same conditions to which the dog alerted the target person. This suggests that there is some level of odour consistency for the same condition across different people and is consistent with studies that reveal odour biomarkers associated with specific conditions [24]. It further demonstrates that, within this sample, dogs were unlikely to learn new target odours from a novel person. Thirty-one out of thirty-three dogs were able to generalise odours associated with their owners’ condition(s) to a new person but were not seemingly expanding this behaviour to a novel health condition in a novel person. These results highlight the need for future empirical work into the odour profile of certain conditions, and across different individuals, to gain further understanding into what these dogs are detecting, and how this may impact their decision making.

But as reported in the results, two dogs were reported to alert other people to conditions other than those that they alerted the target person. For the first dog, the conditions that the dog was reported to alert to other people were injuries (sprains) and muscle spasms. In the case of a dog alerting to injuries, it is difficult to determine whether the dog is in fact alerting to the injury itself or other physiological states associated with the injury. Since this dog also alerted the target person to anxiety, it is possible that the dog was detecting anxiety associated with the injuries and not the injuries themselves. Given that this was the only dog reported to alert to injuries, it is difficult to discuss the phenomenon further. The second dog, however, was reported to alert other people to two different cardiovascular conditions (heart attack and high blood pressure). Again, this dog alerted the target person to anxiety, so it is possible that anxiety played a role in the dog alerting other people to a heart attack and high blood pressure. Furthermore, it is also possible that, for both dogs, the conditions to which they alerted other people were events that elicited attention seeking behaviours from the dogs, and these behaviours were interpreted as alerts. This interpretation of the findings is most likely given that the dogs were unlikely to have reinforcement histories associated with the conditions to which they alerted other people.

Although this study was exploratory in nature, the main limitation of this study was the small sample size which limited the power of any analyses. A further limitation of the study is that the reported results may be impacted by the demographics of the sample; the majority of the sample identified as female and were between the ages of 24–65. Therefore, more generalized comments across MAD owners as a whole cannot be made until further cohort studies have been carried out. In addition, those that responded to the survey may feel more positively about their MAD than those who did not take part in the survey. Indeed, of those that did respond, 81% classified themselves and their MAD as ‘best friends’. It should also be noted that the study was a self-report survey which means that MAD owners reported on their perception of their dogs’ alerting behaviours. Therefore, it is possible that individual and cultural differences in interpreting dogs’ behaviour could have affected the owners’ responses [63] As such, the subjective nature of the study means the findings should be interpreted accordingly. A lack of objectivity likely influences owners’ reports of what their dogs alert to and their perceived level of their dogs’ accuracy [64]. Moreover, the study information sheet presented to participants before they consented to taking part in the study explicitly stated that the researchers were interested in how common it was for dogs that alert to medical conditions to alert to more than one condition or person. Although we stated that we also wanted to hear from dog owners whose dogs did not alert to multiple conditions or people, it is possible that demand characteristics resulted in participants unconsciously misreporting their dogs’ alerting behaviours. While the current study documents a previously anecdotal phenomenon, follow up objective studies are needed to assess the frequency and development of these behaviours. Our results should be interpreted cautiously.

Almost half of respondents stated that their dog had received no formal training for medical alert, and of those that were trained, seventy-seven percent were trained by the owners themselves. It is possible that dogs who have had formal training (e.g., from an accredited training establishment or charity) for a single odour and have an owner that is less likely to reward any alerts outside of the specific criteria, are less likely to alert to multiple conditions or multiple people. However, within this sample, whether the dog was formally trained or not was not significantly associated with whether or not the dog alerted to multiple conditions or multiple people. It should be noted, however, that dogs formally trained from training establishments/charities represented only twenty-three percent of the sample, therefore any differences may only emerge with more participants per group.

This study sought to capture a sample representative of MADs and, as many operational MADs are working without having formal training, it was considered important that all MAD owners were included in the sample. Overall, given the small number of participants representing certain training categories, results pertaining to factors that may impact the likelihood that a dog may, or may not, alert to multiple conditions or people should be taken with caution. However, despite the small sample size, it is clear that many MAD owners report that their dog alerts to multiple conditions, people, or both and that MAD owners perceive their dog to be highly accurate in their alerts.

Conclusion

This study sought to document the phenomenon that dogs alert to multiple health conditions and to multiple people. The results showed that a large majority of MAD owners reported that their dog alerts to multiple conditions, over half of respondents reported that their MAD alerts to multiple people, and just under half of participants reported that their dog alerted to both multiple conditions and multiple people. Dogs were reported to successfully alert to 33 different health conditions, and mostly alerted other people to the same conditions that they alerted the target person. Owners perceived their dogs to be highly accurate with their alerts. The results may suggest overlapping VOC profiles between different conditions, as dogs alerting to multiple health conditions and multiple people could suggest that there are some common odours across conditions to which the dogs are responding. Furthermore, it is possible that the phenomena of dogs’ alerting to multiple conditions and/or multiple people are, through a combination of the dogs’ liberal alerting and the owners’ reward contingencies, shaped over time with the owner. Given the results of other studies which found discrepancies between owner reports and objective assessments of MADs, these phenomena should be assessed directly in future studies. Such studies may wish to objectively investigate the rate at which this phenomenon is occurring, and begin to address how, and why, it is emerging and/or reinforced in working MADs. What is apparent from the results of this study is that many untrained and formally trained dogs are reported to be alerting to multiple health conditions and to multiple people.

The top six most frequently reported conditions dogs alert to and for dogs that alert to those conditions, the other conditions they are also reported to alert.

(DOCX)

Click here for additional data file.

The results of Fishers exact tests for sociodemographic variables of the target person and dog and whether or not the dog alerted to multiple conditions, multiple people, or both.

(DOCX)

Click here for additional data file.

The conditions that dogs alerted other people that were different from the conditions to which the dog alerted the target person.

(DOCX)

Click here for additional data file.

(DOCX)

Click here for additional data file.

(SAV)

Click here for additional data file.

10 Dec 2020

PONE-D-20-32952

Medical Alert Dogs are alerting to multiple conditions and multiple people

PLOS ONE

Dear Dr. Reeve,

Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process.

You will find the detailed reviewer response below. They draw special attention to the statistical analysis and how it is reported in the paper. In the interest of getting you a timely response to your submission, we are not seeking additional statistical reviewer input on this version, but it is likely that we will do so when sending a revised version for review.

Please submit your revised manuscript by Jan 24 2021 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at plosone@plos.org. When you're ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

Please include the following items when submitting your revised manuscript:

A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.

A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.

An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.

If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter.

If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols

We look forward to receiving your revised manuscript.

Kind regards,

I Anna S Olsson, Ph.D.

Academic Editor

PLOS ONE

Journal Requirements:

When submitting your revision, we need you to address these additional requirements.

1. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at

https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main_body.pdf and

https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_title_authors_affiliations.pdf

2. In your Methods section, please provide additional information about the participant recruitment method and the demographic details of your participants. Please ensure you have provided sufficient details to replicate the analyses such as: a) the recruitment date range (month and year), b) a description of any inclusion/exclusion criteria that were applied to participant recruitment, c) a table of relevant demographic details, d) a description of how participants were recruited, and e) descriptions of where participants were recruited and where the research took place.

3. Please include additional information regarding the survey or questionnaire used in the study and ensure that you have provided sufficient details that others could replicate the analyses. For instance, if you developed a questionnaire as part of this study and it is not under a copyright more restrictive than CC-BY, please include a copy as Supporting Information.

4. Please include captions for your Supporting Information files at the end of your manuscript, and update any in-text citations to match accordingly. Please see our Supporting Information guidelines for more information: http://journals.plos.org/plosone/s/supporting-information.

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #1: Partly

**********

2. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: No

**********

3. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: Yes

**********

4. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: Yes

**********

5. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: This paper reports on an exploratory study about Medical Alerting Dogs and their owners, about whether the dogs alert to multiple conditions and to multiple owners. It is an interesting and timely piece of work, that is well written and easy to follow.

I do have a few major concerns:

- The authors document sociodemographic information for MAD, but why did they choose not to examine whether people other than the primary person, if dogs alerted to multiple people/multiple conditions, actually had a condition and which one that was?

- When dogs also alerted to other conditions, it would be interesting to examine which these conditions were. Given that there many conditions that were reported infrequently, I would focus this analysis on the more common ones (anxiety, hypoglycaemia, hyperglycaemia, seizure and POTS). This could be a descriptive analysis, not an inferential one, but it would allow to remove some speculation that currently occurs in the discussion.

- The paper does not report any test statistics, degrees of freedom, or P-values. I would like to see these added to substantiate the claims made by the authors.

- In the statistical analysis, many comparisons have been made. Because of this, significant findings can occur by chance. A more conservative way is to correct for the number of statistical tests done by dividing the cut-off value (P-value) by the number of tests run. The authors did not choose to do this, but it should be revised by a statistician whether it is not more correct to do so still.

There are also some minor concerns:

INTRODUCTION

- Line 46: reference 1 is incomplete in the list of references. Please correct.

- Line 54: in addition to reference 14, there is also the work by Martos Martinez-Caja (2019 in Epilepsy & Behavior), which is more recent, to support this.

- Line 88: I do not know the word ‘brinsel’. I have looked it up, but could not retrieve its meaning. Can this word be substituted, if not explained?

- Line 93: “It is possible that certain aspects of their training”. Replace “their” by “the dogs’ “, so that it is more clear.

- Line 107: typo – “on” should be “an”

MATERIALS AND METHODS

- Lines 120-125 and line 128: in which language was the survey designed? For a UK target audience? Is anything known about the country of origin of the participants?

- Line 129-130: rather than having people ask for the survey, I would suggest to include it in full as supplementary material

- Line 167: To make this sentence easier to read, I would add “for” between “knowledge,” and “what condition(s)”

- Line 187: typo – “altered” should be “alerted”

RESULTS

- Table 2: this is a very important table, that could be even more informative if it was also added how many dogs were trained/untrained when they alert to a particular condition. Also, a table or figure should be able to stand on its own, meaning everything in it must be clear. Therefore, I would suggest explaining the abbreviations POTS and PTSD below the table.

- Line 248: the numbers here (90% and 10%) do not correspond to those in table 5. Please check and adjust as necessary.

- Lines 279-281: I would break this down for the most commonly alerted conditions as well.

DISCUSSION

- Line 316 and further: this discussion could be substantiated if the authors examined their data in a bit more detail for the most commonly reported conditions that dogs alert to (see major comment). At the moment it seems that dogs mainly alert the same condition in other people in the household, but surely that is more common for some conditions than others, since it is unlikely e.g. that two people from the same household have epilepsy. Although I do appreciate the conciseness with which the authors present their results, a more detailed presentation of the data would help here.

- Line 360: “for whether they decide to alert” – I would be careful with this kind of working. A dog does not decide to intentionally alert a person. The dog may decide to intentionally perform a particular behavior (which the owner then interprets as an alert).

- Lines 476-477: also, how was the study presented to the respondents? Is there a risk of response bias in that repondents may have felt they were expected to have a dog that responds to multiple conditions/people. Perhaps the authors could elaborate on this as well?

**********

6. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: No

[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.]

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at figures@plos.org. Please note that Supporting Information files do not need this step.

15 Jan 2021

Dear Dr Olsson,

Thank you for taking the time to review our manuscript and have a reviewer provide feedback. We are pleased to have an opportunity to revise our manuscript. In the revised manuscript, we have carefully considered the reviewers comments and suggestions and have addressed and/or responded to each of their points. The reviewer’s comments were very helpful overall, and we are appreciative of such constructive feedback on our original submission. After addressing the issues raised, we feel the quality of the paper is much improved. Following the reviewer’s suggestions, we have added three additional tables as supplementary materials (file entitled: S1 S2 S3 Tables), the captions for which can be found at the very end of the manuscript file.

Upon closer inspection of the manuscript, we have also made a few further edits (i.e., editing references that were formatted incorrectly). All changes are indicated with track changes in the file named Reeve et al – Revised Manuscript with Track Changes.doc.

Thank you for your time and we look forward to hearing from you.

Sincerely,

Dr Catherine Reeve

Response to Reviewer Comments

Dear Reviewer,

Thank you for taking the time to review our manuscript, Medical alert dogs are alerting to multiple conditions and multiple people. We very much appreciate your feedback. You highlighted a number of ways in which we could improve the quality of our manuscript, and we have made the changes accordingly. Please find our responses to your comments below:

Major Comments

Comment:

The authors document sociodemographic information for MAD, but why did they choose not to examine whether people other than the primary person, if dogs alerted to multiple people/multiple conditions, actually had a condition and which one that was?

Response:

Thank you for your query. We would like to point out that in lines 169-172 (in the revised manuscript without track changes), participants were asked, to the best of their knowledge, what conditions their dog had alerted to other people. We did not ask participants to report specifically what conditions the dogs were alerting to specific people and therefore cannot report this level of detail. But we agree that the findings related to the conditions that dogs alert other people to were not discussed in much detail and, therefore, the conditions to which dogs were reported to alert to other people have now been presented in Table 7.

Furthermore, in lines 291-295 we state that 94% of participants reported that if the dog was alerting to other people, it was alerting the same condition(s) to which the dog alerted the target person. Only 6% (n=2) of those who reported that their dog alerted to other people reported that their dog alerted other people to a different condition. In response to your comment, we have added the details of those who reported that their dog alerted other people to conditions other than those conditions to which the dog alerted the primary person in a table in the supplementary materials (S3) and have discussed these findings further in the discussion (Lines 483-497).

Comment:

When dogs also alerted to other conditions, it would be interesting to examine which these conditions were. Given that there many conditions that were reported infrequently, I would focus this analysis on the more common ones (anxiety, hypoglycaemia, hyperglycaemia, seizure and POTS). This could be a descriptive analysis, not an inferential one, but it would allow to remove some speculation that currently occurs in the discussion.

Response:

You have raised an important point and we agree that knowing which other conditions dogs alert to would be a valuable addition to the manuscript. A supplementary table (S1 Table) has been added that presents the top six most frequently reported conditions that dogs alert to and, for dogs that alert to those conditions, the other conditions they are reported to alert. We have also conducted further analyses to examine whether there are any significant associations between whether or not a dog alerts to each of these conditions and whether or not the dog alerts to multiple conditions (see response to next comment).

Comment:

The paper does not report any test statistics, degrees of freedom, or P-values. I would like to see these added to substantiate the claims made by the authors.

Response:

Thank you for raising an important point. The only inferential statistics performed were Fisher’s exact tests because comparisons were either 2x2 tables, or, for tables larger than 2x2, one or more of the cells in the table contained counts less than 5 (based on the assumptions of Chi-Square tests of independence as discussed by McHugh, 2013). When using Fisher’s exact tests only the p value is reported.

The authors had previously discussed included binomial logistic regression to analyse whether any of the variables in Table 1 predicted whether or not a dog alerted to multiple conditions, multiple people, or both multiple conditions and multiple people, but considering that a basic assumption of regression analyses is that there are roughly 12 participants per independent variables, our small sample size led us to decide that the results of such analyses would be not be valid.

But considering the comments about presenting more detailed analyses of the top six conditions to which dogs were reported to alert, additional analyses were included. We analysed whether alerting to one of the top six conditions was associated with a greater likelihood of alerting to other conditions, other people, or both other conditions and other people. Results revealed that there were no significant relationships between the variables. These results are now reported on lines 238-242, 285-289, and 311-314.

Comment:

In the statistical analysis, many comparisons have been made. Because of this, significant findings can occur by chance. A more conservative way is to correct for the number of statistical tests done by dividing the cut-off value (P-value) by the number of tests run. The authors did not choose to do this, but it should be revised by a statistician whether it is not more correct to do so still.

Response:

Thank you for highlighting this error on our part. We have adjusted our analyses and used a Šidák correction for multiple comparisons. As a result of the correction for multiple comparisons, one of our previously significant findings is no longer significant (association between gender of target person and whether or not the dog alerts to multiple conditions) and as such, discussion of this finding has been removed.

Minor Comments

Introduction

• Line 46: reference 1 is incomplete in the list of references. Please correct.

>> Reference 1 (now on line 558) has been completed.

• Line 54: in addition to reference 14, there is also the work by Martos Martinez-Caja (2019 in Epilepsy & Behavior), which is more recent, to support this.

>> Thank you very much for pointing out this relevant paper. We have included the paper as a citation on line 54 and it is included in the reference list (number 15, line 595).

• Line 88: I do not know the word ‘brinsel’. I have looked it up, but could not retrieve its meaning. Can this word be substituted, if not explained?

>> We appreciate that this term is not one that is commonly used outside of medical detection dog training. We have also identified that it was, in fact, spelled incorrectly. We have therefore edited the spelling of the word accordingly, and have further included a brief definition on lines 89-90.

• Line 93: “It is possible that certain aspects of their training”. Replace “their” by “the dogs’ “, so that it is more clear.

>> The recommended edit has been made on line 94.

• Line 107: typo – “on” should be “an”

>> This error has been corrected on line 108.

Materials and Methods

• Lines 120-125 and line 128: in which language was the survey designed? For a UK target audience? Is anything known about the country of origin of the participants?

>> Thank you for pointing out the lack of detail regarding the survey itself and the participants. The survey was only available in English and available to participants worldwide. We have added this information as well as general descriptive analyses of the locations of participants on lines 122, 126. 203-205.

• Line 129-130: rather than having people ask for the survey, I would suggest to include it in full as supplementary material

>> You have raised an important point however, we feel that it would be best to maintain that the survey be available only upon request, simply because the survey flow is quite complex. The survey was designed to have four major branches depending upon the nature of the participants’ relationship with the medical alert dog and it is therefore a long and complicated-looking document. If the editor and reviewer do not consider the length and flow of the survey to be an issue, we will be happy to provide the survey itself.

• Line 167: To make this sentence easier to read, I would add “for” between “knowledge,” and “what condition(s)”

>> This edit has been made on line 170.

• Line 187: typo – “altered” should be “alerted”

>> This error has been corrected on line 190.

Results

• Table 2: this is a very important table, that could be even more informative if it was also added how many dogs were trained/untrained when they alert to a particular condition.

While we agree that this would be a very interesting addition to the table, unfortunately it is not possible to add this information. When participants with trained MADs were asked which condition their dog was first trained to alert to, they often also included the other conditions to which the dog began alerting in their response. Therefore, we are unable to differentiate between conditions for which the dog was trained and those that it was not trained for but began alerting.

• Also, a table or figure should be able to stand on its own, meaning everything in it must be clear. Therefore, I would suggest explaining the abbreviations POTS and PTSD below the table.

>> Thank you for pointing out this error on our part. Table 2 has been edited accordingly and tables added since review have also included these footnotes (including the supplementary materials).

• Line 248: the numbers here (90% and 10%) do not correspond to those in table 5. Please check and adjust as necessary.

>> The numbers you are referring to are now found in Table 3 and Table 4. In Table 3, we would like to bring your attention to the variable “Dog alerts to multiple conditions”. In this table, the total number of participants that reported that their dog does not alert to multiple conditions is 10. In Table 5, the total number of dogs that are reported to alert to single condition is also 10. Therefore, we are unable to identify any errors in the data between these tables. If we have incorrectly identified your specific area of concern, please do not hesitate to respond accordingly.

• Lines 279-281: I would break this down for the most commonly alerted conditions as well.

>> We agree that this would be a valuable addition to the manuscript. We have therefore added Table 7, which presents the conditions to which dogs were reported to alert other people from most commonly reported to least commonly reported. We also conducted further analyses to examine whether or not dogs that alerted to these top six conditions were associated with whether or not a dog alerted to multiple people (Lines 285-289). Lastly, we also included further information in the supplementary material (S3 Table), which presents the data for the two dogs that were reported to alert other people to conditions other than those conditions they dog alerted to the target person.

Discussion

• Line 316 and further: this discussion could be substantiated if the authors examined their data in a bit more detail for the most commonly reported conditions that dogs alert to (see major comment). At the moment it seems that dogs mainly alert the same condition in other people in the household, but surely that is more common for some conditions than others, since it is unlikely e.g. that two people from the same household have epilepsy. Although I do appreciate the conciseness with which the authors present their results, a more detailed presentation of the data would help here.

We appreciate the reviewer highlighting an important area where we could substantiate our claims. Additional analyses revealed no significant associations between whether dogs alerted anxiety and whether or not they alerted to other conditions, people, or both. But considering the conditions to which dogs were reported to alert to other people, further descriptive analyses revealed that anxiety was the second most common condition that dog alerted other people. We have further examined the most common conditions to which dogs alerted other people and have presented this data in Table 7. As indicated in the table, anxiety was the second most commonly reported condition that dogs alerted to other people. This point was discussed in the discussion at more length.

• Line 360: “for whether they decide to alert” – I would be careful with this kind of working. A dog does not decide to intentionally alert a person. The dog may decide to intentionally perform a particular behavior (which the owner then interprets as an alert).

>> Thank for you highlighting this point. We agree with your statement that dogs do not have a threshold for whether to decide to alert a person, but rather, have a threshold for whether they perform an alert behaviour. We have changed the wording of this sentence accordingly (line 384).

• Lines 476-477: also, how was the study presented to the respondents? Is there a risk of response bias in that respondents may have felt they were expected to have a dog that responds to multiple conditions/people. Perhaps the authors could elaborate on this as well?

>> Thank you for raising an important point. The information sheet presented to participants before completing the survey did in fact state that researchers were interested in whether or not their dog alerted to multiple conditions or multiple people. This could have resulted in demand characteristics influencing participants’ responses. This point has been discussed on lines 508-515.

We hope these changes are suitable. We feel that the revised manuscript is much stronger after making your recommended changes. Thank you for your time,

Dr Catherine Reeve

Animal Welfare and Behaviour

School of Psychology

Queen’s University Belfast

Submitted filename: Reeve et al. - Response to Reviewer Comments.docx

Click here for additional data file.

25 Feb 2021

PONE-D-20-32952R1

Medical Alert Dogs are alerting to multiple conditions and multiple people

PLOS ONE

Dear Dr. Reeve,

Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process.

You will find the reviewer feedback as well as editorial feedback below.

Please submit your revised manuscript by Apr 11 2021 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at plosone@plos.org. When you're ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

Please include the following items when submitting your revised manuscript:

A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.

A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.

An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.

If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter.

If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols

We look forward to receiving your revised manuscript.

Kind regards,

I Anna S Olsson, Ph.D.

Academic Editor

PLOS ONE

Journal Requirements:

Please review your reference list to ensure that it is complete and correct. If you have cited papers that have been retracted, please include the rationale for doing so in the manuscript text, or remove these references and replace them with relevant current references. Any changes to the reference list should be mentioned in the rebuttal letter that accompanies your revised manuscript. If you need to cite a retracted article, indicate the article’s retracted status in the References list and also include a citation and full reference for the retraction notice.

Additional Editor Comments (if provided):

Among the limitations of the study, please add a reflection on that the data are owner-reported, thus strictly speaking it is about owners perceiving their dogs to be alerting to multiple conditions. Given what you write on lines 510-511, "Given the results of other studies which found discrepancies between owner reports and objective assessments of MADs", it is important to mention this as a study limitation.

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation.

Reviewer #1: (No Response)

**********

2. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #1: Yes

**********

3. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: Yes

**********

4. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: Yes

**********

5. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: Yes

**********

6. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: Dear authors,

Thank you very much for the thorough consideration of all my comments. They were sufficiently addressed.

I have just two small remaining remarks (one that may require some revision still, the other is just a comment).

1) Since this survey was only done in English and distributed worldwide via the snowballing technique, how sure are you that the respondents indeed mastered the English language enough to understand your questions? Also, could cultural differences have an impact on the responses to your specific questions? It might be good to add a few lines about this in your discussion.

2) Regarding my previous comment about line 248 in the original document and numbers (10% and 90%) not matching with table 5: unfortunately I did not save a digital copy of your draft, nor did I retain the paper version where I made my first set of comments on. Editorial Manager also does not allow me to retrieve the original manuscript, so I cannot go back and check my own comment. Very sorry about this. But, looking at the current text and tables, it all seems clear now!

I look forward to seeing this paper as a published article!

**********

7. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: No

[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.]

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at figures@plos.org. Please note that Supporting Information files do not need this step.

12 Mar 2021

Dear Dr Olsson,

Thank you and the reviewer for taking the time to review our manuscript. Thank you for taking the time to review our manuscript and have a reviewer provide feedback. We are pleased to have an opportunity to revise our manuscript once more. In the revised manuscript, we have carefully considered both you and the reviewer’s comments and suggestions and have addressed and/or responded to each of the points raised. The feedback was very helpful overall, and we are appreciative of such constructive feedback on our original submission. After addressing the issues raised, we feel the quality of the paper is much improved.

Please note that in our previous submission we neglected to address the comments/edit suggestion made by yourself, the editor. Therefore, we have addressed the previous comments as well as the most recent comments together here.

Response to Editor Comments

Comment:

Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming.

Response:

We have made formatting and style edits to the title page and manuscript according to the PLOs ONE style template.

Comment:

In your Methods section, please provide additional information about the participant recruitment method and the demographic details of your participants. Please ensure you have provided sufficient details to replicate the analyses such as: a) the recruitment date range (month and year), b) a description of any inclusion/exclusion criteria that were applied to participant recruitment, c) a table of relevant demographic details, d) a description of how participants were recruited, and e) descriptions of where participants were recruited and where the research took place.

Response:

A) Recruitment date range was added on Line 124.

B) Inclusion criteria were explicitly identified through the addition of a header on Line 128

C) Demographic details can be found in Table 1.

D) Recruitment details were explicitly identified through the addition of a header on Line 121.

E) Where participants were recruited from was flagged by the reviewer in the previous round of revisions. This information has been added in Line 122. We also clarified that the survey was completed by participants on their own electronic device on Line 177.

Comment:

Please review your reference list to ensure that it is complete and correct. If you have cited papers that have been retracted, please include the rationale for doing so in the manuscript text or remove these references and replace them with relevant current references. Any changes to the reference list should be mentioned in the rebuttal letter that accompanies your revised manuscript. If you need to cite a retracted article, indicate the article’s retracted status in the References list and also include a citation and full reference for the retraction notice.

Response:

We have carefully reviewed all of our references and can confirm that they are accurate and that none have since been retracted. This is further confirmed by the fact that all papers have active DOIs (which have been added to the references as is highlighted in the revised manuscript)

Response to Reviewer Comments

Comment:

Among the limitations of the study, please add a reflection on that the data are owner-reported, thus strictly speaking it is about owners perceiving their dogs to be alerting to multiple conditions. Given what you write on lines 510-511, "Given the results of other studies which found discrepancies between owner reports and objective assessments of MADs", it is important to mention this as a study limitation.

Response:

Thank you for identifying this important point. We have added Lines 509-513 which highlight this aspect of the study and caution the reader to interpret the results accordingly.

Comment:

Since this survey was only done in English and distributed worldwide via the snowballing technique, how sure are you that the respondents indeed mastered the English language enough to understand your questions? Also, could cultural differences have an impact on the responses to your specific questions? It might be good to add a few lines about this in your discussion.

Response:

Regarding the first point, we did not require that participants indicate that they had mastered the English language at any point in the survey. We did require, however, that participants read the participant information sheet and then consent to a series of statement prior to beginning the survey. We presume that anyone who could not comprehend the information sheet and the consent statements would not have consented to participating.

We do not have any reason to believe that participants’ responses to the questions themselves could be affected by cultural differences. We did, however, add a comment about the potential for participants’ interpretation of their dogs’ behaviour to be different across cultures (Line 511), as supported by the addition of another reference (64, Amici et al., 2019).

We hope these changes are suitable. We feel that the revised manuscript is much stronger after making your recommended changes. Thank you for your time,

Dr Catherine Reeve

Animal Welfare and Behaviour

School of Psychology

Queen’s University Belfast

Submitted filename: Reeve at al. - Response to Reviewers.docx

Click here for additional data file.

15 Mar 2021

Medical Alert Dogs are alerting to multiple conditions and multiple people

PONE-D-20-32952R2

Dear Dr. Reeve,

We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements.

Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication.

An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at http://www.editorialmanager.com/pone/, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to-date. If you have any billing related questions, please contact our Author Billing department directly at authorbilling@plos.org.

If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact onepress@plos.org.

Kind regards,

I Anna S Olsson, Ph.D.

Academic Editor

PLOS ONE

Additional Editor Comments (optional):

Reviewers' comments:

22 Mar 2021

PONE-D-20-32952R2

Medical Alert Dogs are alerting to multiple conditions and multiple people

Dear Dr. Reeve:

I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department.

If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact onepress@plos.org.

If we can help with anything else, please email us at plosone@plos.org.

Thank you for submitting your work to PLOS ONE and supporting open access.

Kind regards,

PLOS ONE Editorial Office Staff

on behalf of

Dr. I Anna S Olsson

Academic Editor

PLOS ONE