A randomized controlled trial of animal-assisted activities for pediatric oncology patients: psychosocial and microbial outcomes

Abstract

Introduction:

Evidence about effectiveness and safety of dog visits in pediatric oncology is limited.

Method:

We conducted a randomized controlled trial of volunteer-based dog visits vs usual care among pediatric oncology inpatients. Psychological functioning and microbial load from hand wash samples were evaluated. Parental anxiety was a secondary outcome.

Results:

Among 26 patients, we did not observe a difference in the adjusted mean present functioning score (−3.0; 95% confidence interval [CI]: −12.4, 6.4). The difference in microbial load on intervention vs control hands was −0.04 (95% CI: −0.60, 0.52) log10 CFU/mL, with an upper 95% CI limit below the pre-specified non-inferiority margin. Anxiety was lower in parents of intervention vs control patients.

Discussion:

We did not detect effects of dog visits on patient functioning; however, our study was underpowered due to low recruitment. Visits appeared to improve parental anxiety. With proper hand sanitization, visits did not increase hand microbial levels.

Clinical trial registration:

clinicaltrials.gov NCT03471221

Introduction

Children with cancer may experience many emotions, including distress. Strategies to improve their symptoms, hospital experience, and health-related quality of life are needed (Cheng et al., 2019; Jibb et al., 2022; Linder & Hooke, 2019; Sherman et al., 2006). Visiting dog programs are common in pediatric oncology (Chubak & Hawkes, 2016; Chubak et al., 2022). However, evidence on the safety and effectiveness of animal-assisted interventions (AAI) in pediatric oncology is limited (Cotoc et al., 2019). Theoretical models suggest multiple mechanisms (e.g., distraction, movement, touch, and increased human interaction) through which AAI affect physical and psychological outcomes (Holder et al., 2020). In pediatric populations, AAI generally reduce pain (Braun et al., 2009; Feng et al., 2021; Sobo et al., 2006), provide comfort (Bardill & Hutchinson, 1997; Caprilli & Messeri, 2006; Wu et al., 2002), and have positive emotional effects (Kaminski et al., 2002; McCullough et al., 2018; Mulvaney-Roth et al., 2023; Silva & Osorio, 2018; Vidal et al., 2023).

In the pediatric oncology population specifically, non-randomized studies suggest a benefit of AAI (Chubak et al., 2017; Cotoc et al., 2019; Gagnon et al., 2004; Silva & Osorio, 2018). However, only one randomized controlled trial (RCT) of AAI in this population has been published to date (McCullough et al., 2018). The McCullough et al. study intervention consisted of multiple 10- to 20-minute non-structured sessions (i.e., no prescribed activities), approximately weekly over four months, in an outpatient setting. There were no significant differences in patients’ psychosocial outcomes over time in the intervention group compared to the control group. Despite the scarcity of AAI research in pediatric oncology, AAIs are often available to these patients (Chubak & Hawkes, 2016). Identifying which AAI approaches are effective – and for which outcomes – can help providers and hospitals determine how to best provide supportive care to pediatric oncology patients.

It is also important to understand whether AAI poses safety risks. The McCullough RCT of pediatric oncology patients did not study microbial transfer from animal to child (McCullough et al., 2018). A recent systematic review called for more research on AAI safety (Dalton et al., 2020). Though therapy dogs can harbor pathogens, little is known about whether pathogens are transferred during AAI visits (Dalton et al., 2020). Two recent studies have suggested that microbial transfer can occur, either as part of group visits (i.e., multiple children visiting the dog at the same time) (Dalton et al., 2021) or when children are permitted to give dogs treats directly from their hands (Edner et al., 2021). In a 2014 survey of top pediatric oncology centers, 8 out of 19 centers reported not offering AAI mainly because of concerns about infection (Chubak & Hawkes, 2016). Understanding which microbes are transferred to, and remain on, patients’ hands after standard sanitation will further our understanding of the potential for AAI to cause infections.

The aims of our RCT were, therefore: 1) To assess the effect of therapy dog visits on psychosocial outcomes and satisfaction with care among pediatric oncology inpatients; 2) To determine whether therapy dog visits increase microbial levels on children’s hands. We hypothesized that the intervention would improve patients’ psychosocial state and would not increase the microbial load on their hands.

Methods

Trial design

We conducted a randomized, two-arm, parallel-group trial, allocating patients 1:1 to dog visits (intervention) or usual care (control). The trial (NCT03471221, “Helping Improve Pediatric Patient Outcomes”, HIPPO) was approved by the Seattle Children’s institutional review board (STUDY00000844), to which Kaiser Permanente Washington ceded. The National Institutes of Health (R21HD091877) and Kaiser Permanente funded this study but did not participate in other ways. The study conforms to US Federal Policy for the Protection of Human Subjects standards.

Participants

English-speaking patients 5–17 years old on the inpatient Cancer Care Unit at Seattle Children’s Hospital were eligible irrespective of cancer type, sex, race, or ethnicity. Patients were ineligible if they were allergic to or afraid of dogs, were on the bone marrow transplant service, had isolation precautions in place, had non-intact skin on hands, did not have an English-speaking parent/legal guardian (henceforth, “parent”), or did not have a parent able to provide written consent. The minimum eligible age was initially 6 years; however, 5 months into the study (after 10 patients were enrolled), it was lowered to 5 years to increase enrollment.

Study staff identified potentially eligible patients through medical record review. Parents provided written informed consent and participants provided assent. Up to one parent per child provided written informed consent for the optional parent surveys.

Interventions

The intervention consisted of hospital-based visits (approximately 1/week) from a handler and dog team. Patients could receive up to 4 visits. Study staff reviewed patient records to assess eligibility for the intervention on each day that visiting dog sessions were offered. Seven experienced dog and handler teams participating in the Seattle Children’s Hospital Visiting Dog program conducted the intervention visits.

Individual patient visits occurred in each patient’s room for up to 20 minutes. Anyone who touched the dog was required to sanitize their hands before touching the dog and at the end of the visit. Activities were at the discretion of the patient and the handler and included petting the dog, watching the dog perform a trick, and talking with the handler. Activities followed the hospital’s regulations and procedures for dog visits. Study staff supervised visits.

Since our goal was compare the intervention to what patients would otherwise be doing (Chubak et al., 2021), control arm patients received usual care. Our goal was not to compare the intervention to another specific activity, but rather to compare dog visits to the more general and heterogenous condition of no intervention. During the 20-minute period between measurements (described below), control patients engaged in whatever activities they wished. No study-specific restrictions or suggestions were made. Study staff did not stay in the patient room during this period. Control patients were not required to sanitize their hands, as hand sanitization was considered part of the intervention (Figure 1).

Figure 1:

Timing of data collection at visit 1 in randomized controlled trial of dog visits for pediatric oncology patients

Outcomes

Research staff collected data at baseline (pre-randomization), pre-visit 1 (just before the first intervention/control period, while patients were still blinded), post-visit 1 (just after the first intervention/control period), the day before or of first hospital discharge, 2–3 days after first hospital discharge (follow-up 1), and 9 weeks after first hospital discharge (follow-up 2). Primary outcomes were evaluated post-visit 1 (Table 1 and Figure 1). Primary outcomes were not collected at subsequent visits. Patient and parent participants were each offered $15 after the first follow-up survey and $20 after the final survey.

Table 1.

Timing of study survey instruments for randomized controlled trial of therapy dog visits in pediatric oncology

Instrument	Baseline (before randomization)	Before 1st visit	After 1st visit	Discharge	Follow-up
PedsQL™ Present Functioning Visual Analog Scales (anxiety, sadness, anger, worry, fatigue, and pain at present moment)	C	C	C	-	-
Positive and Negative Affect Schedule (PANAS) short versions for parents	P	-	-	P	P
Hospital rating item from the CAHPS® Child Hospital Survey (modified)	-	-	-	P‡	P‡
Questions about being in the hospital	-	-	-	C and P‡	C and P‡
PedsQL™ Cancer Module (pain, nausea, procedural anxiety, treatment anxiety, cognitive problems, perceived physical appearance, communication)	5–7†: P 8–17†: C	-	-	5–7†: P 8–17†: C	5–7†: P 8–17†: C
Spielberger State-Trait Anxiety Inventory state scale short form	P‡	-	P‡	-	-

Abbreviations: C, child; P, parent/legal guardian; y/o, years old

^†Based on age (in years) at enrollment

^‡Only among parents who consented to participate in the study

Follow-up surveys were either mailed or sent as an email link, depending on participant preference. If no response to the electronic survey was received within 5 days, an automatic reminder email and survey link was sent. Staff followed up with phone calls to those who received mailed surveys. If no response was received after another 5 days, study staff mailed another survey regardless of survey mode preference. Surveys collected at hospital discharge and follow-up are not presented due to low follow-up rates.

During each intervention patient’s first visit, study staff recorded which dog team performed the visits, hand sanitization status, which hands (if any) touched the dog, protocol deviations, visit duration, and who entered the room. During control arm visits, study staff waited outside the room and recorded who entered the room.

Psychosocial outcomes

The primary effectiveness outcome was post-visit 1 patient present functioning measured by the PedsQL^™ Present Functioning Visual Analog Scales (PedsQL VAS) total symptom score (Sherman et al., 2006). The instrument consists of 6 items and instructs respondents to rate “how you feel now.” Respondents write an “X” along a 10 cm horizontal line going from a smiling face to a frowning face with labels such as “not worried” to “very worried.” The distance (in mm) along the line represents the score. The total symptom score is the average of the 6 item scores. Individual items (i.e., anxiety, sadness, anger, worry, fatigue, and pain), the emotional distress summary score (i.e., average of the anxiety, sadness, anger, and worry items), and pre-post changes in scores were secondary or exploratory outcomes.

Secondary outcomes included the Positive and Negative Affect Schedule for Children-Parent report (PANAS-C-P) (Ebesutani et al., 2012), PedsQL^™ Cancer Module (Varni et al., 2002), a modified hospital rating item from the Consumer Assessment of Healthcare Providers and Systems (CAHPS^®) Child Hospital Survey (The CAHPS Child Hospital Survey, 2016), study-specific questions about being in the hospital, and parent anxiety (Spielberger State-Trait Anxiety Inventory [STAI] short-form state scale) (Marteau & Bekker, 1992).

Microbial outcomes

The primary safety outcome of the study was microbial levels on patients’ hands at post-visit 1. Detection of clinically important pathogens (Murthy et al., 2015) (Staphylococcus aureus, Enterococcus faecalis, Enterococcus faecium, Acinetobacter species, Pseudomonas aeruginosa, Enterobacteriaceae species, Clostridium difficile, Aspergillus, Malassezia, and Pasteurella) was a secondary outcome. Patients’ hands were sampled using a hand wash sampling technique (Landers & Dent, 2014); samples were aliquoted, vortexed, and plated (Material S1). Colonies were counted as colony forming units (CFUs) and pathogens identified (Material S1).

Medical record review

At enrollment, study staff abstracted patient demographics and cancer diagnosis from the medical record. Hospitalizations, diagnosed infections, positive clinical cultures, antibiotics, antifungals, and psychosocial care during the follow-up period (58 days after enrollment) were also collected. Infections diagnosed and/or positive clinical cultures were classified as true infections (vs colonization) using definitions (Centers for Disease Control and Prevention).

Adverse events

Injuries, allergic reactions, and breaches of confidentiality were evaluated as potential adverse events (AEs). Potential AEs could be identified from the patient’s clinical team, direct report from the patient/parent, direct observation by clinical staff, or medical record review. A physician independent of the study reviewed all potential AEs and assigned attribution. True AEs were defined as possibly, probably, or definitely related to the study. Serious adverse events (SAEs) were defined as any true AE that met standard criteria for a SAE (U.S. Department of Health and Human Services, 2007). A Data Safety Monitoring Board reviewed AE data.

Sample size

We estimated that 40 patients (20 per group) would provide adequate power to achieve study aims. For the effectiveness aim, we estimated we would have 80% power to detect an absolute mean difference in PedsQL VAS total score between groups of 13.0 (smaller than the change of 14.2 observed in our pilot study) (Chubak et al., 2017). This assumed a standard deviation (SD) of 14.7 based on pilot pre-visit scores and a two-sided type 1 error rate of 0.05.

If the two groups truly had no difference in mean microbial load, we estimated that a sample size of 40 would provide 80% power to detect noninferiority with the following margins: Δ=0.4 log10 CFU/mL if the SD of the outcome was 0.5 and Δ=1.3 log10 CFU/mL if the SD was 1.4 (Chow et al., 2003; Julious, 2004). Calculations assumed a 1-sided type 1 error of 0.025 and a SD of 0.5 to 1.4 based on the literature on hand-cleaning methods in non-pediatric populations (Howard et al., 2014; Kampf et al., 2006; Monistrol et al., 2013), given lack of knowledge of microbial load values for pediatric populations. The attained sample size of 26 was based on challenges recruiting, due in part to the COVID-19 pandemic.

Randomization

Participants were assigned (1:1) to intervention or control using a computer-generated random number list with permuted (variable) blocks of size 2, 4, and 6, stratified by age (≥13 vs <13 years). The allocation sequence was generated by the study biostatistician prior to recruitment and was concealed by the REDCap database. Participants were randomly assigned in REDCap after all baseline data had been collected, at which point the research assistant pressed the key to execute group assignment.

Blinding

Participants did not know their group assignment until after pre-visit 1 data collection. Chart abstraction for infections and other outcomes was completed by the same staff that enrolled patients, supervised visits, and collected patient surveys. Laboratory staff did not have participant identification number or know group assignments. The Principal Investigator did not have access to the study databases or participants’ randomization group assignment until analyses of pre-specified primary and secondary outcomes were complete.

Statistical methods

Aim 1

Following an intent-to-treat (ITT) approach, primary Aim 1 analyses compared PedsQL VAS total scores immediately following visit 1 in patients randomized to the intervention vs control group. Linear regression models included an indicator for the randomized group and were adjusted for age (stratification factor: ≥13 vs <13 years) and the pre-randomization value of the outcome, following our pre-specified analytic plan (Material S2). We calculated 95% confidence intervals (CIs). Secondary outcomes measured immediately after visit 1 were analyzed using the same approach. Additional secondary outcomes measured at discharge and at follow-up time points (Table 1 and Figure 1) were described, among respondents, by intervention group.

Aim 2

Primary Aim 2 analyses applied noninferiority testing (a common paradigm for safety endpoints) (Mauri & D’Agostino, 2017) to compare average microbial loads for intervention and control patients’ hands. Analyses determined if the intervention average exceeded the usual care average by greater than a prespecified “acceptable” margin. Since levels of acceptable microbial load following hand sanitization in a pediatric inpatient population were not available from the literature, we set the margin as one SD of the microbial load in the sample at baseline (pre-randomization).

For the safety aim, because ITT analyses may not be conservative, both ITT and per-protocol analyses were conducted (Mauri & D’Agostino, 2017). We planned to reject the null hypothesis of inferiority only if the intervention was noninferior under both analyses.

ITT analyses compared total microbial load on patients’ hands in the intervention vs control group immediately following visit 1. Per-protocol analyses compared microbial load in patients who received a dog visit to control patients who would have been medically eligible for a dog visit had they been assigned to the intervention group. For both sets of analyses, we modeled the hand-level microbial load (log10 transformed), accounting for dependency of repeated measures within individuals, using generalized estimating equations (Liang & Zeger, 1986). We fitted a linear regression model including the randomization group, age group, and the baseline value of the hand-level microbial load. We applied jackknife standard error estimates (Paik, 1988). We then applied the noninferiority test by constructing a 2-sided 95% CI for the intervention effect and rejecting the null hypothesis if the upper limit of the CI was smaller than the pre-specified margin; this procedure corresponds to a 1-sided type 1 error rate of 0.025.

In a sensitivity analysis, we compared only hands that touched a dog to control patients’ hands. For secondary outcomes of the presence of clinically important organisms, we described the proportion of patients with the outcome by study arm.

Complete case analyses were conducted for patients with available outcome data for each study measure. For all aims, results of statistical tests were interpreted cautiously, recognizing that power to achieve our study aims was limited due to our inability to meet our target sample size. Several pre-planned exploratory aims (Material S2) were not conducted because of either too few events or COVID-19-related delays leaving insufficient time to complete all analyses.

Results

Participation

Recruitment began in March 2018 and ended in February 2020 due to the COVID-19 pandemic. Out of 436 patients assessed for eligibility (Figure 2), we randomized 26 patients: 14 to the control group and 12 to the intervention group. Nineteen parents participated (Table 2).

Figure 2:

Participant flow for randomized controlled trial of dog visits for pediatric oncology patients

*Reasons (not mutually exclusive): Not English-speaking (n=37), no parent available to provide written consent (n=8), skin on hands not intact (n=20), on bone marrow transplant service (n=4), in isolation (n=44), allergic or sensitive to dogs (n=3)

**Reasons: care team not contacted (n=37), care team contacted but did not approach family (n=26), family approached but uninterested (n=14)

***2 control patients were discharged before they were available for visit 1

Table 2.

Baseline characteristics of study participants, by randomization group

Children	Intervention (N=12)	Control (N=14)
Demographic and medical characteristics	n (%)	n (%)
Age (years)
5–12	7 (58.3)	8 (57.1)
13–17	5 (41.7)	6 (42.9)
Mean (SD)	11.7 (4.0)	12.7 (2.7)
Sex
Male	7 (58.3)	8 (57.1)
Female	5 (41.7)	6 (42.9)
Other	0 (0.0)	0 (0.0)
Race
American Indian/Alaska Native	0 (0.0)	2 (14.3)
Asian	0 (0.0)	0 (0.0)
Black/African American	0 (0.0)	0 (0.0)
Native Hawaiian/Pacific Islander	1 (8.3)	0 (0.0)
White/Caucasian	10 (83.3)	8 (57.1)
More than one race	0 (0.0)	0 (0.0)
Unknown or not reported	1 (8.3)	4 (28.6)
Hispanic ethnicity
Hispanic	2 (16.7)	2 (14.3)
Not Hispanic	10 (83.3)	10 (71.4)
Unknown or not reported	0 (0.0)	2 (14.3)
Oncologic diagnosis
Leukemia/lymphoma	6 (50.0)	10 (71.4)
Sarcoma	2 (16.7)	2 (14.3)
Brain	2 (16.7)	1 (7.1)
Neuroblastoma	1 (8.3)	1 (7.1)
Other	1 (8.3)	0 (0.0)
Days between hospital admission and randomization
0–2	4 (33.3)	6 (42.9)
3–14	5 (41.7)	5 (35.7)
≥15	3 (25.0)	3 (21.4)
Mean (SD)	8.8 (10.3)	13.6 (28.2)
Pets at home
Yes	10 (83.3)	12 (85.7)
Dog(s)	8 (66.7)	10 (74.1)
Cat(s)	3 (25.0)	6 (42.9)
Other	3 (25.0)	1 (7.1)
No	2 (16.7)	2 (14.3)
Dominant hand
Left	1 (8.3)	3 (21.4)
Right	11 (91.7)	10 (71.4)
Both	0 (0.0)	1 (7.1)
Psychological functioning	Mean (SD)	Mean (SD)
PedsQL™ Present Functioning Visual Analog Scales†
Total	25 (25)	22 (17)
Emotional distress summary score (anxiety, sadness, anger, worry)	21 (26)	16 (17)
Anxiety	15 (24)	14 (20)
Sadness	14 (19)	19 (27)
Anger	23 (32)	9 (12)
Worry	33 (37)	21 (25)
Fatigue	45 (32)	44 (30)
Pain	19 (29)	25 (27)
PedsQL Cancer Module treatment anxiety†	78 (21)	82 (24)
PedsQL Cancer Module total score†	69 (14)	71(16)
PANAS-C-P positive affect score‡	15 (5)	16 (5)
PANAS-C-P negative affect score†	12 (5)	12 (6)
Hand sampling	n (%)	n (%)
Hand(s) sampled
Right only	0 (0.0)	0 (0.0)
Left only	0 (0.0)	0 (0.0)
Both	12 (100.0)	14 (100.0)
Staphylococcus aureus present§	2 (16.7)	3 (21.4)
Vancomycin-resistant enterococci present	0 (0.0)	0 (0.0)
Malassezia present	0 (0.0)	0 (0.0)
Aspergillus present	0 (0.0)	1 (7.1)
Pasteurella present	0 (0.0)	0 (0.0)
Enterobacteriaceae present§	3 (25.0)	0 (0.0)
Acinetobacter present§	1 (8.3)	1 (7.1)
Pseudomonas aeruginosa present§	0 (0.0)	0 (0.0)
Clostridium difficile present	0 (0.0)	0 (0.0)
	Mean (SD)	Mean (SD)
Total microbial load on right hand (CFU/mL), log10 scale	3.46 (1.21)	3.88 (0.92)
Total microbial load on left hand (CFU/mL), log10 scale	2.93 (1.39)	3.61 (1.18)
Parents	Intervention (N=9)	Control (N=10)
Demographic and medical characteristics	n (%)	n (%)
Gender
Female	8 (88.9)	8 (80.0)
Male	1 (11.1)	2 (20.0)
Non-binary	0 (0.0)	0 (0.0)
Other	0 (0.0)	0 (0.0)
Prefer not to answer	0 (0.0)	0 (0.0)
	Mean (SD)	Mean (SD)
Spielberger State-Trait Anxiety Inventory state scale short-form score†	53.0 (15.4)	45.0 (10.9)

Abbreviations: CFU, colony-forming units; CHG, chlorhexidine gluconate; PANAS-C-P, Positive and Negative Affect Schedule for Children-Parent Report; SD, standard deviation

^†Lower scores indicate better outcomes

^‡Higher scores indicate better outcomes

^§Further evaluated for antimicrobial resistance and no resistant phenotypes identified

The groups had similar baseline PedsQL VAS total scores (Table 2). The emotional distress score was slightly higher (worse) among intervention patients at baseline, as were the anger and worry scores. Most participating parents were female (84%). Parents of intervention group children had, on average, a higher (worse) baseline anxiety score than parents of control group children (Table 2).

All 12 intervention group patients received visit 1. Visits occurred a median of 9 days after hospitalization (interquartile range [IQR]: 2.5–15.5 days). Five children touched the dog with both hands and five with one hand; two received a dog visit but did not touch the dog. All intervention patients sanitized their hands at the beginning and end of the dog visit. One protocol deviation occurred when a dog licked a patient. One participant did not complete the post-visit 1 survey. Six intervention group patients received additional dog visits. Response proportions to discharge, follow-up 1, and follow-up 2 surveys (% with any non-missing outcome) were 50% (6/12), 25% (3/12), and 42% (5/12), respectively.

Of the 14 patients randomized to the control group, 12 (86%) had visit 1 (usual care) and completed the survey and hand sampling. Visits occurred a median of 3 days after hospitalization (IQR: 2–10.5 days). Response proportions to discharge, follow-up 1, and follow-up 2 surveys were 57% (8/14), 50% (7/14), and 43% (6/14) respectively.

Effect of intervention on psychological outcomes

The mean PedsQL VAS scores were similar between intervention and control group patients (Table 3). For the total score (adjusted mean difference −3.0 [95% CI: −12.4 to 6.4]) and all sub-scores except for anxiety, adjusted mean differences were negative (i.e., lower symptoms in the control group than the intervention group). These estimates were consistent with no effect.

Table 3.

Psychological outcomes after first visit among patients and parents randomized to visiting dog intervention vs. control, intention to treat

	Intervention (N analyzed=11)†	Control (N analyzed=12)†	Difference
	Mean (SD)	Mean (SD)	Mean (95% CI)‡
Patients
PedsQL™ Present Functioning
Visual Analog Scales¶
Total score	11.0 (12.5)	13.6 (12.8)	−3.0 (−12.4, 6.4)
Emotional distress summary score	8.2 (14.1)	8.0 (11.4)	−1.6 (−9.8, 6.7)
Anxiety	5.1 (10.3)	4.2 (6.2)	0.4 (−5.5, 6.3)
Sadness	5.3 (8.4)	8.8 (12)	−1.7 (−7.9, 4.5)
Anger	8.6 (20.9)	7.2 (13.6)	−3.2 (−16.7, 10.4)
Worry	13.9 (20.7)	11.8 (17)	−1.3 (−14.5, 11.8)
Fatigue	26.7 (35)	33.6 (26.9)	−5.6 (−32.2, 20.9)
Pain	6.5 (9.5)	15.9 (24.7)	−8.0 (−25.0, 9.0)
Parents	(N analyzed=9)	(N analyzed=9) §
Spielberger State-Trait Anxiety Inventory state scale short-form score¶	38.9 (14.9)	46.3 (11.6)	−13.7 (−21.4, −6.0)

Abbreviations: CI, confidence interval; SD, standard deviation

^†Two out of 14 patients randomized to the control arm did not have a visit 1; one out of 12 patients randomized to the intervention group did not complete the survey

^‡Estimated from a linear regression model adjusted for age group and the baseline score

^§One out of 10 parents of patients randomized to the control group who consented to participate in the study did not complete the survey

^¶Lower scores indicate better outcomes

After visit 1, parents of intervention group children had a lower anxiety level than parents of control group children (adjusted mean difference of −13.7 [95% CI: −21.4 to −6.0] on the STAI short-form state scale, possible range of 20–80) (Table 3).

Low response rates at discharge and follow-up precluded us from modeling differences in outcomes by randomization group status. Descriptive results are presented in supplementary tables and should be interpreted cautiously.

Effect of intervention on hand microbial levels

The adjusted mean difference between intervention and control groups for microbial load on patients’ hands was −0.04 (95% CI: −0.60, 0.52) CFU/mL, log10 scale (Table 4). This difference was small, and the upper 95% CI limit was below the pre-specified non-inferiority margin of 1 SD of the outcome at baseline (1.19 log10 CFU/mL). The per-protocol and ITT estimates were the same because groups did not change. S. aureus was present in cultures from 3 of 12 (25%) control and 1 of 12 (8%) intervention patients’ hands. Acinetobacter was present in one intervention patient’s culture. No other tested pathogen was present.

Table 4.

Microbial load on patient hands after first visit among patients randomized to visiting dog intervention vs. control

	Intervention	Control	Difference
	Mean (SD)	Mean (SD)	Mean (95% CI)†
Total microbial load (CFU/mL) on hand, log10 scale
All hands ‡
Intention to treat	3.03 (1.32)	3.39 (1.01)	−0.04 (−0.60, 0.52)
Per protocol§	-	-	-
Limited to hands that touched dogamong intervention group patients¶	2.74 (1.26)	3.39 (1.01)	−0.19 (−0.83, 0.46)

Abbreviations: CFU, colony-forming units; CI, confidence interval; SD, standard deviation

^†Estimated using generalized estimating equations (GEE) analysis at the hand level, adjusted for baseline and age group; standard errors were calculated using the Jackknife method. Prespecified non-inferiority margin was 1.19 log10 CFU/mL (estimated as 1 standard deviation of baseline levels)

^‡12 intervention patients (24 hands) and 12 control patients (24 hands) [excluding 2 control patients who did not have a visit 1]

^§Per protocol analyses are equivalent to intention-to-treat analyses since all intervention patients received a dog visit and none of the control group patients who had a first visit were medically ineligible for a dog visit

^¶Of the 24 hands of the 12 intervention patients who had a visit, 15 hands (from 10 patients) were included in this analysis; all 24 hands from 12 control group patients who had a visit were included.

Clinical infections

Clinical infections were assessed on all 26 participants. There were 4 of 12 patients (33%) in the intervention group and 3 of 14 patients (21%) in the control group with ≥1 infection during the study period.

Adverse events and withdrawals

There were 2 deaths in the study, both in the intervention group. Neither was attributable to the intervention. These deaths accounted for all study withdrawals except for one patient in the intervention group who withdrew because they did not want to complete surveys.

Discussion

In this RCT, we did not detect an effect of dog visits on patient functioning as measured by the PedsQL VAS. However, the study closed due to the COVID-19 pandemic before recruitment targets were met. While our study was underpowered to detect our hypothesized effect on patient functioning, our results suggest a reduction in parental anxiety. Furthermore, we found that significant microbial transfer did not occur with therapy dog visits, which included hand sanitization as part of the intervention. It is important to note that AAI interventions vary widely in practice and that our findings from this study might not generalize to different AAI interventions (e.g., that have different hygiene and sanitization procedures) or other pediatric oncology populations (e.g., younger patients)

To date, one other RCT has investigated the effect of AAI on pediatric oncology patients (McCullough et al., 2018). In contrast to our study, McCullough included outpatients, and the intervention consisted of weekly visits over a four-month period. Though McCullough et al. met recruitment targets, the study did not detect an intervention effect on either patient or parental quality of life or anxiety. The authors hypothesized that the lack of effect might have been due to the heterogeneity of their population, which included patients who may not have been as likely to benefit (e.g., older children) (McCullough et al., 2018). Other factors, including insufficient dose, choice of measurement, or the unstructured nature of the intervention, could also have impacted effectiveness.

To our knowledge, ours is the first RCT to examine microbial transfer during individual patient AAI visits in pediatric oncology. In a non-randomized study, Dalton et al. analyzed microbial transfer during pediatric oncology group visits (Dalton et al., 2021). Their findings, based on results from nasal swabs, suggested that microbial transfer occurred and was decreased by applying topical chlorhexidine to dogs’ coats. Another recent study examined microbial transfer from dogs to pediatric oncology patients with two different dogs (Edner et al., 2021). During visits with one of the dogs, children (N=10) were permitted to give the dog treats from their hands. During visits with the other dog, children (N=10) were allowed to give the dog treats only using a pellet feeder and the dog was trained not to lick. Microbial transfer from dogs to children’s fingertips occurred only during visits with the first dog, leading the authors to conclude that allowing patients to give dogs treats directly was the main cause of the transfer. However, other factors, including the specific dog and location of visit, could have been confounders.

A main strength of our study was its RCT design. Having a control group permitted estimation of an intervention effect compared to usual care. Randomization guarded against bias due to confounding. The study’s primary limitation was low enrollment. Recruitment proceeded more slowly than anticipated due to fewer eligible patients than expected and was stopped due to the COVID-19 pandemic. We also did not have prior knowledge of a “safe” microbial load for a pediatric population; thus, a data-driven approach was used to select the margin based on baseline data. Furthermore, we did not conduct analyses of microbial transfer to children’s nares, dogs, or the hospital environment. We also did not power the study to compare clinical infection rates between groups.

In summary, we were not able to detect an effect of therapy dog visits in our study, possibly due to lower power than planned. However, the intervention may have reduced parental anxiety. Future studies of AAI in pediatric settings should also evaluate effects on parents. In addition, our study suggests that microbial load on patients’ hands did not increase after dog visits that included hand sanitization. Large, well-powered studies should examine the effects of AAI, including on infection risk, to identify programs that are both effective and safe.

Data Sharing Statement:

Individual participant data will not be made available due to confidentiality.