Advertisement

Canine detection of volatile organic compounds unique to human epileptic seizure

Open AccessPublished:December 23, 2020DOI:https://doi.org/10.1016/j.yebeh.2020.107690

      Highlights

      • Volatile Organic Compounds unique to human seizure were identified by canines.
      • Canines distinguished between ictal and interictal sweat with 93.7% probability.
      • Non-epileptic seizures were not associated with the unique seizure scent.
      • Canine detection of seizure scent preceded clinical seizure with 82.2% probability.
      • Average duration of warning before an epileptic seizure is 68.2 minutes.

      Summary

      Objective

      Literature accounts of service dogs alerting patients prior to their seizures are a mix of historically poor quality data and confounding diagnoses. In a group of epilepsy patients, Canine Assistants and Florida International University characterized a unique scent combination of volatile organic compounds present during the immediate postictal period, but never at other times. The current study was designed to confirm prospectively if this unique scent, and potential biomarker, can: (1) be detected in an epilepsy monitoring unit (EMU), (2) whether this scent is present with nonepileptic seizures, and (3) whether this scent also precedes the clinical-electrographic seizure.

      Methods

      Following consent and approval, sweat samples taken from EMU admissions at Denver Health Medical Center were sent to Canine Assistants in Georgia. Their team of service dogs, who had been imprinted to recognize the unique scent, were then asked to process these sweat samples in a simple yes/no identification paradigm.

      Results

      Sixty unique subjects were enrolled over a two-year period. In the first part of this study, a total of 298 ictal sweat samples of 680 total observations were collected. The dogs had a 93.7% (OR: 14.89, 95% CI: 9.27, 23.90) probability of correctly distinguishing between ictal and interictal sweat samples. In the nonepileptic seizure population, 18 of the 19 NES events that were accompanied by sweat sample collections were not associated with identification of the unique seizure scent. In the second part of the study, in which subjects had samples collected every hour, dogs identified the unique seizure scent presence before 78.7% of all seizures captured, at a probability of 82.2% (OR: 4.60, 95% CI: 0.98, 21.69) of a positive detection predicting a seizure. The average duration of the warning phase of the scent was 68.2 min. The average duration of the tail phase of the scent faded after 81 min.

      Significance

      This study confirms the unique seizure scent identified by Canine Assistants and FIU may be collected and recognized by dogs trained to do so, in a prospective manner. A significant number of seizures appear to be associated with the unique scent presence prior to clinical-electrical onset of the seizure itself, and therefore further study of this biomarker is warranted.

      Keywords

      1. Introduction

      The unpredictable nature of seizure occurrence is the primary source of morbidity and mortality in epilepsy [
      • Sperling M.R.
      The consequences of uncontrolled epilepsy.
      ]. However, emerging evidence from long-term recording of intracranial electroencephalography (EEG) suggests that seizures do not happen without warning, but instead, changes resulting in a clinical seizure become increasingly more apparent over time [
      • Witte H.
      • Iasemidis L.D.
      • Litt B.
      Special issue on epileptic seizure prediction.
      ,
      • Le Van Q.M.
      • Martinerie J.
      • Navarro V.
      • Boon P.
      • D'Havé M.
      • Adam C.
      • et al.
      Anticipation of epileptic seizures from standard EEG recordings.
      ]. This window of opportunity may offer the possibility of seizure prediction as well as the hope for seizure prevention [
      • Iasemidis L.D.
      Epileptic seizure prediction and control.
      ]. However, implantation of intracranial EEG is not a practical solution for most persons with epilepsy (PWE), and the results from previous attempts at scalp EEG prediction remain wanting [
      • Carney P.R.
      • Myers S.
      • Geyer J.D.
      Seizure prediction: methods.
      ].
      Patients are also interested in seizure prediction and have reported the ability of their personal pets and service animals to “detect” their seizure onset, anecdotally anywhere from 10 s to 5 h in advance, with an average range of 35 min [
      • Strong V.
      • Brown S.W.
      • Walker R.
      Seizure-alert dogs- fact or fiction?.
      ,
      • Strong V.
      • Brown S.W.
      • Huyton M.
      • Coyle H.
      Effect of trained Seizure Alert Dogs on frequency of tonic-clonic seizures.
      ,
      • Burton A.
      What about a seizure-alert dog?.
      ,
      • Kirton A.
      • Wirrell E.
      • Zhang J.
      • Hamiwka L.
      Seizure-alerting and response behaviors in dogs living with epileptic children.
      ]. Physicians and scientists have hypothesized that animals may be cueing into the patient’s behavior, or detecting a change in mood, or a change in odor. Unfortunately, the literature accounts of this cueing signal have been confounded by patients with psychogenic nonepileptic seizures (PNES) [
      • Litt B.
      • Krieger A.
      Of seizure prediction, statistics, and dogs: a cautionary tail.
      ,
      • Krauss G.L.
      • Choi J.S.
      • Lesser R.P.
      Pseudoseizure dogs.
      ,
      • Doherty M.J.
      • Haltiner A.M.
      Wag the dog: Skepticism on seizure alert canines.
      ]. In fact a report of 2 dogs observed during an Epilepsy Monitoring Unit (EMU) admission suggested that in one patient, the dog’s alerting prior to the event may have actually promoted increased PNES frequency [
      • Ortiz R.
      • Liporace J.
      Seizure alert dogs: observations from an inpatient video/EEG unit.
      ]. Other literature reports are either patient testimonies or very small case series that can only demonstrate improved patient quality of life, but lack sufficient data to confirm seizure alerting ability [
      • Kirton A.
      • Winter A.
      • Wirrell E.
      • Snead O.C.
      Seizure response dogs: evaluation of a formal training program.
      ,
      • Brown S.W.
      • Goldstein L.H.
      Can seizure-alert dogs predict seizures?.
      ,
      • Dalziel D.J.
      • Uthman B.M.
      • McGorray S.P.
      • Reep R.L.
      Seizure-alert dogs: a review and preliminary study.
      ]. Canines have been used reliably to signal identification of specific scents for detection of cancer [
      • Moser E.
      • McCulloch M.
      Canine scent detection of human cancers: a review of methods and accuracy.
      ,
      • Cornu J.N.
      • Cancel-Tassin G.
      • Ondet V.
      • Girardet C.
      • Cussenot O.
      Olfactory detection of prostate cancer by dogs sniffing urine: a step forward in early diagnosis.
      ], bombs [
      • Furton K.G.
      • Myers L.J.
      The scientific foundation and efficacy of the use of canines as chemical detectors for explosives.
      ], drugs [
      • Jezierski T.
      • Adamkiewicz E.
      • Sobczynka M.
      • Górecka-Bruzda A.
      • Ensminger J.
      • Papet E.
      • et al.
      Efficacy of drug detection by fully-trained police dogs varies by breed, training level, type of drug and search environment.
      ], and human remains [
      • Oesterhelweg L.
      • Krober S.
      • Willhoft J.
      • Braun C.
      • Thies N.
      • Püschel K.
      • et al.
      Cadaver dogs - a study on detection of contaminated carpet squares.
      ]; however, exploration to identify a specific unique scent as the trigger for seizure-alerting in dogs had not been formally studied.
      In 2016, Canine Assistants (CA; a service dog education program, headquartered in Alpharetta, Georgia) collaborated with the graduate chemistry program at Florida International University (FIU) to explore the possibility of identifying a unique seizure scent signature [
      • Davis P.
      The investigation of human scent from epileptic patients for the identification of a biomarker for epileptic seizures [PhD dissertation on the internet].
      ] responsible for the frequent patient accounts of their dogs’ ability to warn the owner. If identified, could this odor be taught to future CA program graduates to respond to a pending seizure? By collecting samples from saliva, sweat, and breath from 11 EMU-confirmed PWE identified through the Epilepsy Foundation of Florida, their gas-chromatography/mass-spectrometry study identified 11 volatile organic compounds (VOC) from these 495 samples that were only present when collected during the ictal time period and not during interictal time periods. Of these 11, the FIU laboratory believed that 3 of the compounds existed in sufficient concentration to be reasonable candidates for study as a unique human postictal seizure scent signature.
      In 2017, Catala et al. confirmed the ability of dogs trained to distinguish an ictal scent from different body odors of the same person in different contexts and common to different persons with epilepsy [
      • Catala A.
      • Grandgeorge M.
      • Schaff J.L.
      • Cousillas H.
      • Hausberger M.
      • Cattet J.
      Dogs demonstrate the existence of an epileptic seizure odour in humans.
      ]. The sensitivity and specificity obtained were among the highest to date for discrimination of any disease, concluding that seizures are associated with olfactory characteristics.
      This current study extends the original work of the Canine Assistants and Florida International University collaboration, by attempting to confirm the presence of the unique seizure scent in an adult EMU. We aimed to (1) prospectively confirm that seizure scent-sensitized dogs can accurately discriminate an epileptic seizure sample from an interictal sample, (2) assess whether nonepileptic seizures (NES) are also associated with the unique seizure scent, and (3) investigate whether or not the unique seizure scent precedes seizure onset, or only exists in the ictal/postictal time period.

      2. Methods

      2.1 Study population

      The study population was derived from adult EMU patients attended by the principal investigator (EM) admitted for purposes of seizure focus localization or spell characterization at Denver Health’s Brooke Gordon Comprehensive Epilepsy Center, a National Association of Epilepsy Centers Level 4 Center in Denver, Colorado. Patient aged between 18 and 89 years were included. Study consent was obtained under the protocol approved by our institutional review board (COMIRB #18-0399). Spell characterization could produce psychogenic or physiologic nonepileptic seizures/events and are hereafter referred to as NES.

      2.2 Canine teaching method

      A single trainer (JA) was responsible for teaching a team of 13 dogs at the Canine Assistants headquarters in Alpharetta, Georgia. The FIU laboratory produced the 3 VOC mixture (ratio unknown) captured on a sterile cotton pad (100% cotton, 2 × 2, 8 ply, gauze pads; DUKAL Corporation, Syosset, NY, USA), placed in airtight, ten-ml glass, clear, screw top headspace vials with PTFE/Silicone septa (SUPELCO, Bellefonte, PA, USA) and mailed to Canine Assistants.
      At Canine Assistants, from their earliest rearing, all of their service dogs are taught to communicate with their human handlers by responding to yes/no questions. Initially a trainer extends her LEFT hand that has been rubbed with a treat and repeats the word “yes”; followed by extending the unscented RIGHT hand and repeating the word “no”. When the dog goes to smell the LEFT hand or touches its nose to the LEFT hand, the dog is given the treat. Within a few repetitions the dog is able to associate “yes” with touching the LEFT hand, and “no” with touching the RIGHT hand, and over time this becomes generalizable by the dog to a variety of questions presented in this manner [

      Arnold J. Teaching Your Dog to Say Yes or No: The Art of Non-Training. Available from: https://www.psychologytoday.com/us/blog/through-dog-s-eyes/201506/teaching-your-dog-say-yes-or-no-the-art-non-training. Accessed date: 25 March 2020.

      ].
      In this study, after opening the FIU vial, each dog was sequentially presented with the pure standard, and reinforced with the phrase “this is your seizure” and “this is your smell”.
      Testing began by asking each dog “is this your seizure?” and “is this your smell?” while presenting the “seizure” scent or a volunteer’s non-seizure sweat sample. After one or two trials, each dog correctly signaled “yes” (touching of the trainer’s LEFT hand) when the standard “seizure” scent was presented or “no” (touching of the trainer’s RIGHT hand) when a volunteer’s sweat sample was presented. The glass vial was recapped between each presentation, and the entire training was completed within 30 min to ensure as high a VOC concentration as possible for purposes of learning and testing. During the course of the study, not all 13 dogs were used at once. Typical dog teams consisted of 3–4 dogs at a time, but each subject was processed by the same dog team. Conveniently, EMU admissions typically ranged from Monday to Friday so one dog team was used per week of the study. However, between-subject comparisons of each dog were not possible because each dog did not test each subject.

      2.3 Data collection and testing method

      In the Epilepsy Monitoring Unit of the Denver Health Medical Center (Denver, Colorado, USA), sweat samples were collected using a Puritan Sterile Foam Tipped Applicator (6″ plastic handle, blue cap, 1 swab per tube, Guilford, ME, USA). Interictal samples were either collected by rubbing the foam tip applicator in the subject’s axilla for 10 s or by suspending the foam tip applicator and directing subjects to rub the foam tip between their palms for 10 s, prior to carefully placing the swab back into transport tube. The interictal sample collection choice was provided because the FIU laboratory originally screened sweat from the palm only, but given the higher likelihood of obtaining VOCs from axillary sources, both options were offered. Ictal samples were collected by rubbing the foam tip applicator in the subject’s axilla for 10 s. After the swab was replaced into its transport tube, it was marked with the subjects’ initials, date, and time of collection. The EEG study was also digitally marked at the time of the sample collection for later review.
      Study Part 1: Sample collection schedule of phase I EMU patient data (scalp EEG). Sweat sample collections occurred randomly throughout the day and night, usually when neurodiagnostic technology personnel would check on patients during electrode maintenance. Additional sample collections occurred during the care of patients while responding to an ongoing seizure. These subjects are referred to as “low sample density” or “low density” subjects.
      Study Part 2: Sample collection schedule of phase II EMU patient data (patients for whom intracranial EEG had been surgically and/or stereotactically implanted). Sweat sample collections occurred every hour during their admission, at the onset of a clinical seizure, and then every 15 min after a seizure for an additional 2 h. These subjects are referred to as “high sample density” or “high density” subjects.
      All samples for a given subject were gathered at the end of the EMU admission (average duration 3.8 days for phase I subjects and 7.2 days for phase II subjects). Each individual swab was removed from the transport tube, the foam tip was separated from the applicator, and touchlessly dropped into a Mylar Smell Proof Barrier Bag (Premium Vials; 1 oz; 3 1/8″ × 5 1/8″ airtight; zipper stand up pouch; heat seal-able Zip Lok; matte black), and then heat-sealed. For each subject an Excel database table with all samples chronologically ordered by date/time was created for the Georgia testing team. Samples were overnight-shipped to Canine Assistants (Alpharetta, Georgia, USA) for processing by the above canine protocol. In order to minimize the chance that one dog’s response might influence the response of another, each dog processed all the samples in a given shipment one at a time. Because exposing the samples for analysis would likely result in degradation of the odor, only 3–4 dogs were used at any one time, and a positive response from one was counted as a positive response from the testing team. Over the course of the multi-year study, we identified no need for freezing or daily shipping of scent samples by following this method. The results generated a table of yes/no responses per sample and participating dog. These blinded canine results were then later merged with review of EMU discharge reports, combining the date and time of all identified clinical and electrographic seizures or NES.

      2.4 Statistical method

      After all samples were processed by the canine team and EMU-confirmed clinical and electrographic seizures were correlated by date/time stamp, the data were segregated into two groups: ictal observations were assumed to be any sample collected within +/− 90 min of an EMU-documented clinical or electrographic seizure onset (inclusive of NES), and interictal observations were assumed to be any sample collected beyond this 180-min window around all clinical or electrographic seizures (inclusive of NES). This 180-min window was a best-guess assumption based on the literature about when a seizure scent might become present to how long after a seizure the scent might remain [
      • Strong V.
      • Brown S.W.
      • Walker R.
      Seizure-alert dogs- fact or fiction?.
      ,
      • Strong V.
      • Brown S.W.
      • Huyton M.
      • Coyle H.
      Effect of trained Seizure Alert Dogs on frequency of tonic-clonic seizures.
      ,
      • Burton A.
      What about a seizure-alert dog?.
      ,
      • Kirton A.
      • Wirrell E.
      • Zhang J.
      • Hamiwka L.
      Seizure-alerting and response behaviors in dogs living with epileptic children.
      ]. Since epileptic seizures and NES are electrographically and pathophysiologically unrelated disorders, it was also assumed that a NES event sample would not be associated with the presence of the unique seizure scent.
      Subjects were subdivided dichotomously into a group who experienced epileptic seizures, and a group that did not experience epileptic seizures (i.e. physiologic nonepileptic seizures or psychogenic nonepileptic seizures). Those who had epileptic seizures were analyzed to assess the dogs’ ability to detect a difference between ictal and interictal samples. A two-by-two table was derived from the following outcomes: A canine positive-detection of ictal samples are considered true positive, and of interictal samples, false positive. Likewise, a canine negative-detection of interictal samples or NES are considered true negative, and of ictal samples, false negative. Missing or invalid data were considered missing and not included in the analysis. Observations with missing dog detection, or dog smells that did not provide one of the binary answers, and observations with a seizure outcome of UNKNOWN were excluded from the analysis. Sixty-two samples were rejected applying these rules. To determine the probability of the dogs detecting an epileptic seizure correctly a generalized linear mixed model with a logit link was fit with multiple responses for each subject. Within-subject correlation was assumed to have a compound symmetric relationship. Initially, age and sex were included in the model; however, neither was significantly associated with seizure event and, therefore, removed from the final model. Analyses were conducted using SAS software, Version 9.4 (Cary, NC, USA).

      3. Results

      3.1 Subject characteristics

      From 10/9/2018 to 8/5/2020, 60 unique subjects provided consent according to our local institutional review board protocol (COMIRB # 18-0399). In Part 1 of the study, 55 phase I EMU subjects were enrolled. In Part 2 of the study, ten phase II EMU (or “high density”) subjects were enrolled. During the study duration, five phase I subjects from Part 1 of the study went on to have phase II EMU evaluations and therefore also participated in Part 2 of this study. See subject demographics in Table 1.
      Table 1Demographic Information for all patients involved in study.
      CharacteristicsN (%)
      NN60
      Age (Years)Mean (SD)39.5 (14.0)
      Median (IQR)38.5 (27.5, 50.5)
      GenderMale25 (41.7)
      Female35 (58.3)
      RaceAmerican Indian or Alaska Native1 (1.7)
      Asian Indian2 (3.3)
      Black or African American6 (10.0)
      White or Caucasian49 (81.7)
      Other1 (1.7)
      Decline to answer1 (1.7)
      EthnicityNot Hispanic, Latino/a, or Spanish Origin44 (73.3)
      Hispanic, Latino/a, or Spanish Origin12 (20.0)
      Mexican, Mexican American, or Chicano/a3 (5.0)
      Decline to Answer1 (1.7)
      Anti-Seizure Drug Utilization03
      18
      225
      311
      4+13
      Average ASDs2.54
      Seizure Type confirmed by EEGUnique Subjects
      Focal43
      Generalized5
      Of the fifty-five phase I subjects, twelve subjects had or were suspected of having nonepileptic seizures (NES). Of these, seven subjects actually had events during the EMU stay and were formally diagnosed to be NES-only. In Part 1 of the study, thirty-six of fifty-five subjects had at least one valid sweat sample while also having at least one epileptic seizure captured. In Part 2, nine of the ten subjects had at least one valid sweat sample while also having at least one epileptic seizure captured. The average number of observations for Part 1, or “low density” subjects was four. The average number of observations for Part 2, or “high density” subjects was forty-six.
      A total of 680 observations were included in the study; 641 (94.3%) ictal and interictal observations from subjects with epileptic seizures and 39 (5.7%) ictal and interictal observations from subjects with nonepileptic seizures. Of the 641 epilepsy-related observations, 298 (46.5%) fell within the 180-min “ictal” window previously defined, and 343 (53.5%) were defined as “interictal” observations.

      3.2 Aim #1: Prospectively confirm that seizure scent-sensitized dogs can accurately discriminate an epileptic seizure sample from an interictal sample

      Tabled 1
      Seizure Presence YesSeizure Presence No
      Dog Detected Yes19340
      Dog Detected No105303

      3.3 Aim #2: Assess whether nonepileptic seizures (NES) are also associated with the unique seizure scent.

      Seven of the twelve subjects with suspected NES had at least one captured and documented NES event during their EMU admission. Of the 39 total samples attributed to these 12 subjects, 19 were collected during an NES event and within the defined 180 min “ictal” window. Eighteen of these 19 observations (94.7%) were canine negative-detections.

      3.4 Aim#3: Investigate whether or not the unique seizure scent precedes seizure onset, or only exists in the ictal/postictal time period

      Tabled 1
      Seizure Presence YesSeizure Presence No
      Dog Detected Yes16124
      Dog Detected No64241
      A predictive or preictal canine positive-detection was observed in 59 of the 75 seizures captured (78.7%), taken from 106 samples collected within 90 min before an actual seizure event occurred. Of these samples, using a generalized linear mixed model for repeated measurements within subjects, the probability the dog detected the seizure scent prior to the seizure is 82.2% (OR: 4.60, 95% CI: 0.98, 21.69). The warning range of a positive scent detection before a seizure varied from 6 to 177 min (average duration 68.2 min). The duration of the postictal scent presence lasted from 9 to 123 min following the seizure offset (average duration 81 min).

      4. Discussion

      4.1 Seizure thresholds

      Volatile organic compounds are a class of carbon-based chemical compounds with a high vapor pressure at room temperature. These rapidly escaping molecules are released through breath, saliva, and skin (as sweat) and can be produced through a variety of chemical, metabolic and pathological processes [
      • Davis P.
      The investigation of human scent from epileptic patients for the identification of a biomarker for epileptic seizures [PhD dissertation on the internet].
      ]. Sweat from these glands is derived from extracellular fluid, which itself is a reflection of blood plasma chemistry [
      • Davis P.
      The investigation of human scent from epileptic patients for the identification of a biomarker for epileptic seizures [PhD dissertation on the internet].
      ]. Apocrine sweat specifically is the primary source of underarm odor, and apocrine gland release is triggered by fear or sexual arousal [
      • Davis P.
      The investigation of human scent from epileptic patients for the identification of a biomarker for epileptic seizures [PhD dissertation on the internet].
      ]. This relationship between the limbic system and epileptic networks, which frequently involve autonomic signaling, may be responsible for producing the premonitory changes in apocrine release that were observed.
      This signal must reside at some upstream location in the electro-clinical to ictal evolution, since the majority of electrographic seizures were preceded by the seizure scent. While the original hypothesized seizure scent window of +/− 90 min was based on the available literature, our study shows this preictal window may extend beyond that time frame (average warning was 68.2 min but an outlier was as far out as 177 min). Even more compelling, this scent was detectable in both focal onset and generalized onset seizure subjects (See Table 1), suggesting this upstream signal was common to both epilepsy classifications. Epilepsy monitoring is rarely needed to distinguish between focal and generalized epilepsies and our data reflect this reality (and study limitation) by the inclusion of only five subjects with a generalized epilepsy diagnosis. For obvious reasons, subjects with genetic generalized epilepsy were also not included in the phase II cohort, so timing of the seizure scent presence was not possible for this subgroup. Future confirmatory studies should attempt to include a higher density of sample collection (every 15 min), an expanded ictal window informed by these study results, a larger number NES subjects, and larger number of generalized epilepsy subjects to verify these study results in these subpopulations.
      At our institution, phase II patients also receive epilepsy network mapping through direct cortical stimulation of all EEG electrode contacts. Step-wise escalation of electrical current is provided through an Ojemann Cortical Stimulator (Integra; Burlington, MA) to identify seizure threshold of the various EEG contact locations as well as identify symptoms associated with the patient’s seizure as it spreads through the hypothesized network. New or habitual seizures may be triggered through this procedure, but typically seizure symptom mapping is achieved without provoking a seizure. In our cohort of ten phase II subjects, five had sweat samples collected around the time of stimulation. As would be expected, none were associated with a preceding warning scent detection (because cortical stimulation is a provoking factor), but positive scent detections followed initiation of cortical stimulation in these subjects and ranged from 45 to 120 min (average duration 63.5 min). The dog response to this provoked scent appeared to be identical to the CA/FIU scent, but its duration tended to be shorter compared with the subjects’ habitual seizure.
      These clues appear to point to the very beginning of the ictal cascade, near a “seizure threshold” mechanism. A hypothesized limbic/autonomic-associated gating mechanism could be evaluated by modulating parasympathetic tone with a Vagus Nerve Stimulator. Unfortunately at our institution following phase II electrode implantation, anti-seizure drugs and VNS are abruptly stopped to encourage seizures for purposes of epileptogenic zone identification. But patients with a VNS could offer an opportunity to investigate this contribution more thoroughly in a future study.
      Examination of the raw data from one particular phase II subject highlighted another interesting observation (See Table 2). In the excerpted table, the dogs responded negatively every hour beginning at 01:30, but by 05:30 the dogs became unsure (denoted by a question mark). Author JA reported that a strong yes or no was signaled by the dog immediately approaching the appropriate hand. But when a dog walked around in circles after being presented a test odor, it spent longer making a decision when it was unsure. From 08:30 to 11:30 the dogs once again were very confident in their positive response, but then became doubtful from 12:30 to 13:30 and finally resolutely negative again starting at 14:30 and beyond. Despite having intracranial EEG (in the now known seizure focus), no clinical or electrographic seizure was appreciated during this very precise window. Post hoc review of the raw EEG during this time confirmed the initial prospective report (no electrographic seizure was identified), however observation of the spectral analysis during this epoch revealed increased organization (5–7 Hz) that dissipated as the scent faded. This finding apparently reinforces the upstream location of this warning signal and suggests it may not rely on the electroclinical seizure from developing at all. It should be commented that for purposes of statistical analysis, these subject responses were considered “false positive” because they lacked EEG correlate; however, intuitively this pattern of a gradually building then tapering scent is not only compelling for an ictal event but also consistent with more typical electrographic seizures seen in other subjects during the study.
      Table 2Phase II subject with no clinical or electrographic seizure during stated time period.
      DateTimeDog ResponseEMU Documented Seizure
      8/29/20191:30NoNo event
      8/29/20192:30NoNo event
      8/29/20193:30NoNo event
      8/29/20194:30NoNo event
      8/29/20195:30?No event
      8/29/20196:30?No event
      8/29/20197:30?No event
      8/29/20198:30YesNo event
      8/29/20199:30YesNo event
      8/29/201910:30YesNo event
      8/29/201911:30YesNo event
      8/29/201912:30?No event
      8/29/201913:30?No event
      8/29/201914:30NoNo event
      8/29/201915:30NoNo event
      8/29/201916:30NoNo event
      8/29/201917:30NoNo event

      4.2 Pitfalls

      A long history of skepticism surrounds the topic of canine detection of seizures. From literature accounts of seizure-alert dogs provoking psychogenic NES to the often mythical status pets and service animals can have in our patients, controlling for confounding variables is critical to understand the magnitude of any perceived effect of scent dog studies in general.
      To address the concern for bias in these animal studies, Johnen et al. conducted a meta-analysis of available peer-reviewed scent dog studies to develop recommendations for a best practice standard: (1) ideally the target odor should be distinct and chemically known rather than variable, (2) the perfect scent detection task should be a clear yes/no decision for every sample provided and paradigm should use a variable number of target and nontarget samples, (3) storage and handling should be aimed at minimizing cross contamination and loss of target odor, (4) the handler must be blinded in addition to the animal, (5) trainer/handler must be experienced, and (6) type of dog should be listed, but no specific breed recommendations were possible [
      • Johnen D.
      • Heuwieser W.
      • Fischer-Tenhagen C.
      An approach to identify bias in scent detection dog testing.
      ].

      4.2.1 Best practice standard 1

      This study builds on the collaboration between Canine Assistants and the Florida International University. Dr. Kenneth Furton is a leading scholar in forensic chemistry, specializing in scent detection. He founded FIU’s International Forensic Research Institute (IFRI) and continues to direct graduate student research. In his analytical chemistry PhD dissertation, Philip Davis employed instrumental techniques to first compare volatile organic compound (VOC) profiles of samples taken from epileptic individuals to the VOC profiles taken from healthy individuals, and secondly to compare VOC profiles of samples taken from epileptic individuals immediately following a seizure event compared with their baseline samples taken in the absence of seizure activity. His novel work involved discovery of the optimal parameters to maximize human epilepsy scent VOC identification through headspace solid phase microextraction combined with gas chromatography-mass spectrometry. Hand odor, saliva, and breath were analyzed for each of the above sample groups. Fig. 1 shows results of the entire VOC profile for hand samples collected from both groups in the absence of seizure activity. Through scatterplot comparisons of each sample source and between groups, the FIU laboratory sequentially isolated three VOCs that were unique to postictal VOC profiles from hand, saliva, and breath (see Fig. 2).
      Figure thumbnail gr1
      Fig. 1Comparison of hand odor VOC profiles from persons with epilepsy and healthy individuals.
      Figure thumbnail gr2
      Fig. 2Scatterplot comparisons of each sample source (hand, saliva, and breath) and between groups (healthy controls, persons with epilepsy), isolating three VOCs that were unique to postictal VOC profiles.
      Once identified, canine confirmation commenced. First, the produced odor was imprinted in four Canine Assistants dogs. Second, these dogs correctly identified the odor 5 consecutive times from a group of samples containing one blank gauze, one gauze kept in a glass jar for 6 h, one gauze with the scent of the tester (JA), and one interictal sample from an epilepsy patient. Third, the dogs were presented a postictal sample from the same epilepsy patient above, and all dogs confirmed presence of the odor. Fourth, three new dogs were imprinted on the same postictal sample from the epilepsy patient above. Finally, the three new dogs were presented the original produced VOC sample, and all three confirmed presence of the actual postictal sample odor.

      4.2.2 Best practice standard 2

      Described in the methods, author JA utilized a yes/no paradigm not only for imprinting the unique seizure scent but also for testing of all samples shipped from Denver. Because seizures occurred randomly in the Epilepsy Monitoring Unit, differing numbers of samples were shipped each week. To reduce testing fatigue of the dogs not all interictal samples were shipped. Limiting the total number of samples per week to less than 100 also helped to increase variability of the ratio of target to nontarget scents.

      4.2.3 Best practice standard 3

      Described in the methods, our protocol adopted best practices recommended by the FIU laboratory. In the Catala study (not available at the time of our study design) samples were frozen at the collection site in Flavigny sur Moselle, France and then refrozen on arrival in Indianapolis, Indiana until the time of processing [
      • Catala A.
      • Grandgeorge M.
      • Schaff J.L.
      • Cousillas H.
      • Hausberger M.
      • Cattet J.
      Dogs demonstrate the existence of an epileptic seizure odour in humans.
      ]. Retrospectively, we do not feel that freezing is necessary given our ability to achieve interpretable results with the protocol previously described.

      4.2.4 Best practice standard 4

      Double blinding was achieved by separation of the data collection site from the data processing site. The principal investigator (EM) was responsible for interpretation of EEG data, but all samples were collected and packed for shipping only by the study coordinator from the Colorado site (KN). This was designed to increase separation of the knowledge of when seizures occurred, but also to provide a “standardized” source of scent contamination which was felt to be unavoidable due to package handling. At the data processing site in Georgia the dog handler (JA) was provided each week’s list of samples marked by subject initials, date, and time, but no details of when seizures occurred. Responses of yes/no were recorded for each sample of each subject, then both data sets were fused by the study coordinator (KN), further maintaining blinding of the principal investigator (EM).

      4.2.5 Best practice standard 5

      Author JA is Founder and Executive Director of Canine Assistants, a nonprofit organization that teaches and provides service dogs for children and adults with physical disabilities or special needs since 1991. In 2012, she developed the Bond-Based Approach philosophy to their service dog education program which emphasizes improved communication and bidirectional connection through mutual trust, respect, and confidence. This method differs from traditional canine training programs that emphasize task performance and obedience to commands. While this process is lengthier than traditional programs, CA has placed almost 3000 service dogs since 1991.

      4.2.6 Best practice standard 6

      Please see Table 3.
      Table 3Canine Team Information.
      NameAgeSexBreed
      Annabelle6FGolden
      Butch12MGolden doodle
      Catelyn4FGolden
      Cersci4FGolden
      Great3FGolden doodle
      Haddie5FGolden
      Juliette5FGolden doodle
      Luna4FGolden
      Merry3MGolden
      Roch5MGolden doodle
      Stuart6MLabrador
      Vanna3FGolden
      Wally4MGolden doodle

      4.3 Implications for nonepileptic seizure characterization

      While all epilepsy patients would benefit from having a 45–60-min warning before their seizure, it also became evident that the lack of a seizure scent could be useful for characterization of nonepileptic seizures in the EMU. Unfortunately not powered to definitively demonstrate this, observationally, 18/19 (94.7%) NES samples were negative for the scent. Future studies could confirm that a scent-sensitized therapy dog could shorten characterization EMU stays with corroborating video, EEG, and sweat analysis.

      4.4 Identity of the Canine Assistants/Florida International University postictal scent

      The specific chemical identity of the postictal seizure scent is of obvious interest to the field of epilepsy scent detection research. Compounds found only in the postictal odor profiles of epileptic patients include: menthone, menthyl acetate, 3-ethoxy-3,7-dimethyl-1,6-octadiene, camphor, pentadecanal, valencene, (−)-β-bourbene, β-cubene, and 4-tert-butylcyclohexyl acetate. It is unclear what function these substances perform in the clinical scenario, but clear understanding of their function is also beyond the scope of the current study. What we believe we have demonstrated is the first proof of concept that the postictal scent discovered by Canine Assistants and Florida International University is actually produced in the preictal time period and may allow a measurable and intervenable biomarker for human epileptic seizure.

      5. Conclusion

      Our study confirmed the postictal scent identified by Canine Assistants and Florida International University, as potentially unique to epilepsy, was identifiable prospectively in persons with epilepsy during their seizure, and not between seizures. Additionally this unique epileptic seizure scent was not present during nonepileptic seizures. Finally, and perhaps most importantly, the epilepsy seizure scent appeared to precede the majority of EEG/EMU captured seizures by a considerable amount of time. Sixty-eight minutes of warning would allow many therapeutic or logistical safety interventions.

      Acknowledgments

      I wish to thank Brooke Gordon and the generous Gordon Family for their financial support of this exciting research endeavor. I also wish to thank Jennifer Arnold and Darlene Perales of Canine Assistants for multiple trips to Denver to help streamline our study protocol, as well as work their dogs through the hundreds of samples we mailed over the last 2 years. And special thanks to Great, Merry, Butch, Annabelle, Catelyn, Cersci, Haddie, Juliette, Luna, Roch, Stuart, Vanna and Wally. They all deserve a steak each.

      Disclosures

      None of the authors has any conflict of interest to disclose. We confirm that we have read the Journal’s position on issues involved in ethical publication and affirm that this report is consistent with those guidelines.

      References

        • Sperling M.R.
        The consequences of uncontrolled epilepsy.
        CNS Spectr. 2004; 9: 98-109
        • Witte H.
        • Iasemidis L.D.
        • Litt B.
        Special issue on epileptic seizure prediction.
        IEEE Trans Biomed Eng. 2003; 50: 537-539
        • Le Van Q.M.
        • Martinerie J.
        • Navarro V.
        • Boon P.
        • D'Havé M.
        • Adam C.
        • et al.
        Anticipation of epileptic seizures from standard EEG recordings.
        Lancet. 2001; 357: 183-188
        • Iasemidis L.D.
        Epileptic seizure prediction and control.
        IEEE Trans Biomed Eng. 2003; 50: 549-558
        • Carney P.R.
        • Myers S.
        • Geyer J.D.
        Seizure prediction: methods.
        Epilepsy Behav. 2011; 22: S94-S101
        • Strong V.
        • Brown S.W.
        • Walker R.
        Seizure-alert dogs- fact or fiction?.
        Seizure. 1999; 8: 62-65
        • Strong V.
        • Brown S.W.
        • Huyton M.
        • Coyle H.
        Effect of trained Seizure Alert Dogs on frequency of tonic-clonic seizures.
        Seizure. 2002; 11: 402-405
        • Burton A.
        What about a seizure-alert dog?.
        Lancet Neurol. 2017; 16: 265-266
        • Kirton A.
        • Wirrell E.
        • Zhang J.
        • Hamiwka L.
        Seizure-alerting and response behaviors in dogs living with epileptic children.
        Neurology. 2004; 62: 2303-2305
        • Litt B.
        • Krieger A.
        Of seizure prediction, statistics, and dogs: a cautionary tail.
        Neurology. 2007; 68: 250-251
        • Krauss G.L.
        • Choi J.S.
        • Lesser R.P.
        Pseudoseizure dogs.
        Neurology. 2007; 68: 308
        • Doherty M.J.
        • Haltiner A.M.
        Wag the dog: Skepticism on seizure alert canines.
        Neurology. 2007; 68: 309
        • Ortiz R.
        • Liporace J.
        Seizure alert dogs: observations from an inpatient video/EEG unit.
        Epilepsy Behav. 2005; 6: 620-622
        • Kirton A.
        • Winter A.
        • Wirrell E.
        • Snead O.C.
        Seizure response dogs: evaluation of a formal training program.
        Epilepsy Behav. 2008; 13: 499-504
        • Brown S.W.
        • Goldstein L.H.
        Can seizure-alert dogs predict seizures?.
        Epilepsy Res. 2011; 97: 236-242
        • Dalziel D.J.
        • Uthman B.M.
        • McGorray S.P.
        • Reep R.L.
        Seizure-alert dogs: a review and preliminary study.
        Seizure. 2003; 12: 115-120
        • Moser E.
        • McCulloch M.
        Canine scent detection of human cancers: a review of methods and accuracy.
        J Vet Behav. 2010; 5: 145-152
        • Cornu J.N.
        • Cancel-Tassin G.
        • Ondet V.
        • Girardet C.
        • Cussenot O.
        Olfactory detection of prostate cancer by dogs sniffing urine: a step forward in early diagnosis.
        Eur Urol. 2011; 59: 197-201
        • Furton K.G.
        • Myers L.J.
        The scientific foundation and efficacy of the use of canines as chemical detectors for explosives.
        Talanta. 2001; 54: 487-500
        • Jezierski T.
        • Adamkiewicz E.
        • Sobczynka M.
        • Górecka-Bruzda A.
        • Ensminger J.
        • Papet E.
        • et al.
        Efficacy of drug detection by fully-trained police dogs varies by breed, training level, type of drug and search environment.
        Forensic Sci Int. 2014; 237C: 112-118
        • Oesterhelweg L.
        • Krober S.
        • Willhoft J.
        • Braun C.
        • Thies N.
        • Püschel K.
        • et al.
        Cadaver dogs - a study on detection of contaminated carpet squares.
        Forensic Sci Int. 2008; 174: 35-39
        • Davis P.
        The investigation of human scent from epileptic patients for the identification of a biomarker for epileptic seizures [PhD dissertation on the internet].
        Florida International University, 2017 (Available from: https://search.proquest.com/docview/2103114969)
        • Catala A.
        • Grandgeorge M.
        • Schaff J.L.
        • Cousillas H.
        • Hausberger M.
        • Cattet J.
        Dogs demonstrate the existence of an epileptic seizure odour in humans.
        Nat Sci Rep. 2019; 9: 4103
      1. Arnold J. Teaching Your Dog to Say Yes or No: The Art of Non-Training. Available from: https://www.psychologytoday.com/us/blog/through-dog-s-eyes/201506/teaching-your-dog-say-yes-or-no-the-art-non-training. Accessed date: 25 March 2020.

        • Johnen D.
        • Heuwieser W.
        • Fischer-Tenhagen C.
        An approach to identify bias in scent detection dog testing.
        Appl Animal Behav Sci. 2017; 189: 1-12