Can dogs sense schizophrenia

Detection of Impending Aggressive Outbursts in Patients with Psychiatric Disorders: Violence Clues from Dogs

Participants

Patients participating in this study were inpatients of the Dual Diagnosis Ward and the Forensic Psychiatry Inpatient Unit, Abarbanel Mental Health Center, Bat Yam, Israel. The study was approved by the Institutional Review Board (IRB) of Abarbanel Mental Health Center, Israel, Trial Registration 920090532, that is authorised and under the supervision of the Israel Ministry of Health (in accordance with the IACUC guidelines). Data were analysed anonymously. The study was conducted from November 2009 to June 2011.

The wards accommodate an average of 30 adult male patients (age 18+) at any one time, of which 25 are from the Dual Diagnosis Ward. Inclusion criteria: subjects were clinically diagnosed by the heads of wards as having psychotic disorders according to psychiatric evaluations of Structured Clinical Interview for Diagnostic and Statistical Manual of Mental Disorders (DSM)-4-TR (SCID). Of the Dual Diagnosis Ward: 75% of the patients were diagnosed with schizophrenia (co-occurring with substance use disorder), 810% with manic-depressive disorder and 1315% with substance induced-psychosis and a personality disorder. Of the Forensic Psychiatry Inpatient Unit: 80% of the patients were diagnosed with schizophrenia and co-occurring substance use disorder and 20% with personality disorder.

The study protocol maintained the patients standard clinical setting including standard psychiatric medications, psychotherapies etc. Information regarding the patients was obtained from the patients medical files and from the clinical staff. The Israel Ministry of Health approved the IRB Committee of Abarbanel Hospital decision that the dogs were to be considered as a Bio-Medical Device. The dogs were held with a leash by their trainer and owner, Mr Bakeman, at all times.

Mr Bakeman provided his professional trainer certificate and provided evidence for the beneficial effects of alert dogs on patients (e.g. diabetics). The IRB Committee categorised this study as an observational study that involves intervention in the subjects environment. Accordingly, the IRB concluded that there was no need to require a written informed consent from each patient and member of staff, but each would provide a verbal consent.

A record of verbal consent was not required. For the Trial Registration 920090532 the IRB Committee of Abarbanel Hospital and the Israel Ministry of Health approved this consent procedure. Yet, the IRB ordered that for this study the Principal Medical Supervisor, Dr Clara Moray Schild (M.D.), would supervise actions to explain to the patients and staff about the study. This was done as a group during the weekly meetings the patients take in the department (by Dr Delayahu who has been supervised by Dr Schild) or to each individual patient (by Dr Schild). Dr Schild explained to the patients that the aim of the research is to identify violent acts and to prevent violence between patients. Although this explanation may have affected patient behaviour, violent acts still occurred. According to the decision of the IRB Committee, a patient who asked not to be included in the study was provided by Dr Schild with an alternative solution for the time the dogs were at the department.

Patient daily behaviour report

A daily update was received from the medical staff regarding the mental state and behaviour of selected patients in the ward. We also analysed the hospital reports for the number of physical restraint events during years 2010 and 2011. However, as not all cases ended with a physical restraint, this was recorded only as a relative index of violent activity.

Control. The observers filled out an independent report at each session which documented: (1) visualised unusual/violent behaviour of patients; (2) a complementary report on behaviour of patients according to the hospital reports. The observer collected the hospital reports for up to 24hours after each session; the reports were forwarded directly to the principal investigator of this study.

Violent activity questionnaires

A hospital report was completed by the medical staff for each act of violence a patient exhibited. The report described the outburst details including suspected cause, time, severity, medical treatment outcomes. These reports were filled out daily. TableS1 outlines categories for violent outbursts according to nature and severity. The cut-off period for collection of the information was 24hours after a session: (1) This was based on a previous study that recorded positive alerts even 3hours before onset of clinical signs⁶, suggesting alert can occur at least a few hours before the event; (2) Our hypothesis was that every alert is a true alert. This means that alerts which were not followed by relatively immediate outbursts had to be followed; (3) The hospital reports (see below) were provided twice a day (morning and evening). The sessions were conducted at different hours along the day according to availability of the trainer and dogs. In order to standardise the data collection we selected a 24hours cycle; (4) Extending this period also allowed to accommodate the possibility that different mechanisms may be involved in detecting the outbursts, for example: smelling volatiles, which are assumed to be released closely to the time of the event, but, may also be evoked many hours before the event; detecting brain waves, which may precede olfactory cues and, in theory, could be evoked by the patients many hours before events commence. Though studying the methods of detections was not in the scope of this study, the selected settings will promote studying these cues in the future.

Please see Supplementary Information for a full description of Methods of Training Programme of the dogs and Behaviour questionnaire and evaluation of dogs performance

Briefly:

Step 1: Two Golden Retriever litter mate female pups were selected according to the Tests of Campbell method^33,34.

Step 2: At the age of 6+weeks, the pups underwent behavioural analysis tests and initial training to determine their ability for showing general alerting responses.

Step 3: From the age of 2+ months the animals underwent a 30 day obedience training program, during which preliminary obedience skills (common commands: sit, down, stay, halt) were instilled through an operant conditioning method of positive reinforcement reward (pet/food/happy tones).

Step 4: In addition, for this study, the dogs underwent response training such as training for specific barks on command and activation of responses to emergency situations.

At the hospital: Once the dogs reached the age of two months they were also introduced to the hospital inpatient psychiatric ward for further training. At this point, the dogs were trained to bark when a patient outburst occurred. Importantly, in practice, when this training step was successfully completed, the dogs started to alert on impending events (see Results and TableS4- most cases were alerted 30minutes prior to event).

The experiment began when both Golden Retriever dogs were introduced to the hospital ward from age 2 months up to 1.5 years and again at age 2 years. The number of observation days per month was similar between the two periods of the study (Table2).

Their trainer was present at the ward for each session. Most sessions lasted 4hours and only one session was conducted per day, after each session the dogs and owner left the hospital ground. Two dogs were used, each dog worked for 45min alternating periods separated by periods of 1hour and 15min to rest and access to water.

The trainer sat on a chair during each work period (45min) so that the trainer and the dog both had the same visual access to the patients, they were not separated from the patients and potentially could detect auditory and olfactory signals from the patients (Fig.1). Also the patients had visual access to the dogs but, though petting was not allowed, we cannot exclude the possibility that the dogs somehow influenced the behaviour of the patients. In many cases the dogs could not visualise subtle facial movements (cues) of the trainer. Furthermore, most outbursts occurred some minutes and even hours after the alerts (see TableS4), suggesting that in these cases the trainer would not be in a position to transmit specific cues to the dogs.

In addition, our report identified whether the alert was directed at a specific patient or not; these alerts were termed focused and unfocused alerts, respectively.

Video film of the alert dogs

Video recording of the dogs behaviour was approved by the IRB Committee of Abarbanel Hospital. Further to the ethical statements above, the IRB Committee of Abarbanel Hospital approved the final edited video for publication without a written or verbal consent under Trial Registration 920090532. According to the IRB Committee decision, the face of each patient was blurred before allowing the video to leave the hospital. Dr Schild approved that copy and Dr Anavi-Goffer further edited the video. The final version was approved for publication by the IRB Committee of Abarbanel Hospital. Author UB, dog handler, approved the video with a written consent. The video is available via Supplementary Information.

A video film was taken at dog adulthood to demonstrate the dogs behaviour following/at the time of different types of alerts, their behaviour at periods with no alerts and the interaction of the patients with the dogs (petting etc.). The video film suggests that the baseline behaviour of the dogs in the same environment without outbursts was laying on the floor, very often in a resting position. Note: There is a gap of 10min between the real time vs. the time noted on the video. The video time was 10minutes faster than the actual time.

Physical restraints vs. violent outbursts

The wards recorded an average of 0.990.095 (SEM, n=24) physical restraints per day during the years 20102011 (Fig.6A,B). As our trial records were on an hourly basis, dividing by 24, we calculated the mean of the number of physical restraints per hour. In our study, a total of 66 outbursts were recorded in the ward during the first period of study, when the dogs were at pre-puberty, with 304 observation hours. Mean violent outbursts per hour was 0.22 (0.170.27, 95% CI). Similarly, when the dogs were adults, the mean of violent outbursts per hour was 0.17 (0.070.32, 95% CI). These means of violent outbursts per hour were not significantly different. There was no access to the hospital reports apart for the time the dogs worked. Therefore, we could not calculate the mean of general violent outbursts per hour.

Number of outbursts vs. alerts

At dog pre-puberty, 66 violent outbursts were recorded, of which 56 outbursts were alerted by the dogs (True Positive, TP) and 10 outbursts were missed, i.e. the dogs did not alert true outbursts (False Negative; FN) (TableS2). There were 9 alerts without recorded outbursts, i.e. there were no violent outbursts up to 24hours after changes in dogs behaviour were recorded (i.e. False Positive; FP). At dog adulthood, 7 violent outbursts were recorded, of which 6 events were alerted by the dogs (TP) and 1 event was missed (FN). There were no alerts without recorded outbursts. According to the hospital, an average of 30 male patients at any one time in the ward was the basis for calculating no-outburst/no-alert (i.e. ward was fully occupied but during the sessions patients were brought in, or taken out of, the ward to court, psychiatric evaluations, family meetings, etc). The frequency of outbursts was not similar among the patients. The True Negative (TN) cases, i.e. the number of predicted events in which there was no-outburst/no-alert was calculated as follows: based on the hospital reports (Fig.5) we assumed that each patient may develop one violent outburst per each session (=30 events, number of sessions is detailed in Table2), we then subtracted out of 30 events the true positive (TP) cases, false-positive (FP) cases and false-negative (FN) cases of that session. For example the number of True Negative (TN) cases in a session with a dog alert which resulted in a true outburst in the ward (TP), and a dog alert but no outburst (FP), and an outburst in the ward without a dog alert (FN), was calculated as 27 TN cases. The number of TN cases from all sessions was summed. This calculation yielded a final value of 2206 TN cases during dog pre-puberty (TableS2, Fig.2A). In a similar way to above, we calculated 286 TN casesduring dog adulthood(Fig. 2B). Both focused and unfocused alerts and outbursts inside the ward were considered for this analysis (see definitions for focused vs. unfocused alerts at the Results section).

Canine cortisol level

Samples of dog saliva were collected before a session once arriving to the hospital ground (18 samples, labelled baseline) or 2030min after alert within the ward (6 samples, labelled alert) and placed in the freezer of the hospital. Out of all samples, there were 3 pairs of samples (before/after alert) per each dog, a total of 6 pairs. The time point was selected based on a previous report showing that after stress, salivary cortisol level peak after this time point²⁴. Samples were collected between June to September 2010. At the laboratory, samples were prepared according to the manufacturers instructions and assayed for of level of cortisol in saliva using Cortisol Saliva ELISA kit (SA E-6000, Labor Diagnostika Nord). The assay was performed in duplicate and the absorbance was read at 450nm (Infinite, Tecan).

Data integration and statistical analyses

The principal investigators integrated the data. Within the session, data were collected for any alert by each of the dogs. In parallel, data were collected for any outburst of the patients, monitoring the patients continued up to 24hours following each session. The results of patients violent outbursts by dogs were analysed with GraphPad Prism (versions 57) using contingency 22 table with the following categorical variables: the number of correct alert/outburst cases (True Positive (TP) cases), the number of no alert/no outburst cases (True Negative (TP) cases, calculated as detailed above), the number of no alert/outburst cases (False negative (FN) cases) and the number of alert/no outburst cases (false positive (FP) cases). Chi square (pre-puberty) or Fishers exact test (adulthood) analyses, two-sided, with 95% confidence interval (CI), were used to compare categorical variables. In contingency analysis, n is the number of total frequency. Data were analysed anonymously and the patients ID altered during sessions (i.e. subjects were not necessarily repeatedly measured). The results of effect size were analysed with GraphPad Prism (version 7). All hypotheses were tested with two tails, p value<0.05 was considered statistically significant.

The effectiveness of the dog as an alert under repeated observations per animal was analysed using the Cochrans Q test, IBM SPSS version 25. The test examined whether alert rates (alert/no alert) changed over time (hours). Each dog was regarded as a case with repeated measures (over the hours). For each dog, data were analysed from 152hours in pre-puberty (total 304hours for both dogs) and 18hours in adulthood (total 36hours for both dogs), see Table1 (n is 170hours per dog). For this analysis, repeated alerts were regarded as alert only once and both focused and unfocused alerts were analysed. IBM SPSS treated alert as a success. p value<0.05 was considered statistically significant.

Results of ELISA assay for salivary cortisol level were analysed using the One phase decay analysis, GraphPad Prism version 6. The sensitivity of the kit was 0.024ng/ml (0.07nM). One sample (baseline group) was below this level and therefore was excluded. Data were analysed with GraphPad Prism version 8. All samples were analysed using unpaired two-tailed Students t-test. Pairs of samples were analysed by two-way analyses of variance (ANOVA), each column represented a different outcome (baseline or alert) so matched values are spread across a raw, each raw represented a different day, so matched values are stacked into a subcolumn (by a dog). Data are expressed as meansSEM. p<0.05 considered statistically significant.

Mean physical restraints of patients per hour was calculated from raw data and analysed by column statistics (GraphPad Prism). Mean outbursts of patients (in ward) per hour was calculated from raw data and analysed by column statistics (GraphPad Prism). Results are expressed the meansupper/lower CI, p<0.05 considered statistically significant.

At each age, ratio of focused or unfocused dog alerts (numerator) out of total alerts (denominator) and the 95% CI of ratios were calculated using the Wald method (http://www.graphpad.com/quickcalcs/confInterval2/). Data are expressed as ratios95% CI. p<0.05 considered statistically significant.