Author's Accepted Manuscript Highly-Trained Dogs’ Olfactory System Detects Prostate Cancer in Urine Samples Gianluigi Taverna, Lorenzo Tidu, Fabio Grizzi, Valter Torri, Alberto Mandressi, Paolo Sardella, Giuseppe La Torre, Giampiero Cocciolone, Mauro Seveso, Guido Giusti, Rodolfo Hurle, Armando Santoro, Pierpaolo Graziotti
PII: DOI: Reference:
S0022-5347(14)04573-X 10.1016/j.juro.2014.09.099 JURO 11844
To appear in: The Journal of Urology Accepted Date: 19 September 2014 Please cite this article as: Taverna G, Tidu L, Grizzi F, Torri V, Mandressi A, Sardella P, La Torre G, Cocciolone G, Seveso M, Giusti G, Hurle R, Santoro A, Graziotti P, Highly-Trained Dogs’ Olfactory System Detects Prostate Cancer in Urine Samples, The Journal of Urology® (2014), doi: 10.1016/ j.juro.2014.09.099. DISCLAIMER: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our subscribers we are providing this early version of the article. The paper will be copy edited and typeset, and proof will be reviewed before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to The Journal pertain.
Embargo Policy All article content is under embargo until uncorrected proof of the article becomes available online. We will provide journalists and editors with full-text copies of the articles in question prior to the embargo date so that stories can be adequately researched and written. The standard embargo time is 12:01 AM ET on that date. Questions regarding embargo should be directed to [email protected].
ACCEPTED MANUSCRIPT Highly-Trained Dogs’ Olfactory System Detects Prostate Cancer in Urine Samples
Gianluigi Taverna1, Lorenzo Tidu2, Fabio Grizzi1, Valter Torri3, Alberto Mandressi1, Paolo Sardella2, Giuseppe La Torre2, Giampiero Cocciolone2, Mauro Seveso1, Guido Giusti1, Rodolfo Hurle1, Armando
1Humanitas
RI PT
Santoro1, Pierpaolo Graziotti1
Clinical and Research Center, Rozzano, Milan, Italy; 2Italian Ministry of Defense’s, Military
Veterinary Center, CEMIVET, Grosseto, Italy; 3IRCCS Istituto di Ricerche Farmacologiche “Mario Negri”,
M AN U
SC
Milan, Italy.
Notes: This study was partially presented in the Podium Session at the Annual Meeting of the
TE D
American Urological Association, May 16-21, 2014.
Correspondence to: Gianluigi Taverna, Head of Prostate Diseases Section, Department of Urology, Humanitas Clinical and Research Center, Via Manzoni 56 - Rozzano – Milan, Italy Phone: +39 02
AC C
EP
82244535; Fax: +39 02 82244790; E-mail: [email protected]
1
ACCEPTED MANUSCRIPT Abstract Purpose: To establish the diagnostic accuracy, in term of sensitivity and specificity at which a rigorously trained canine olfactory system can recognize prostate cancer (PC)-specific volatile organic compounds (VOCs) in urine samples. Materials and Methods: Two three-year old female German Shepherd Explosive Detection Dogs were trained to identify PC-specific VOCs in urine samples and tested on 902 subjects (362 with PC ranging
RI PT
from very-low risk to metastatic and 540 healthy, affected by non-neoplastic diseases or non-prostatic tumors control participants). This cross sectional design for diagnostic accuracy involved one large Italian teaching hospital and the Italian Ministry of Defense’s, Military Veterinary Center.
Results: The dogs achieved the following performances: Dog 1: sensitivity 100% (95%CI: 99.0-
SC
100.0%) and specificity 98.7% (95%CI: 97.3-99.5%). Dog 2: sensitivity 98.6% (95%CI: 96.8-99.6%) and specificity 97.6% (95%CI: 95.9-98.7%). When only the adult men in the Control Group were considered, Dog 1 achieved a sensitivity of 100% and specificity of 98% (95%CI: 96-99.2%), while Dog
M AN U
2 a sensitivity of 98.6% (95%CI: 96.8-99.6%) and specificity of 96.4% (95%CI: 93.9-98.1%). Analysis of false positive cases did not reveal any consistent pattern in terms of participant demographics or tumor characteristics.
Conclusions: A trained canine olfactory system detects PC-specific VOCs in urine samples with high estimated sensitivity and specificity. Further studies will be necessary to investigate the potential
TE D
predictive value of using this procedure for recognizing PC.
Keywords: Prostate; Cancer; Dog Olfactory System; Volatile Organic Compounds; Sensitivity;
AC C
EP
Specificity.
2
ACCEPTED MANUSCRIPT Introduction Prostate cancer (PC) represents the fifth most frequent cancer in the world 1. Although the prostate specific antigen (PSA) blood testing has increased PC detection, the main drawback is its lack of specificity and accuracy. High serum PSA levels can be detected in men with non-malignant conditions 2.
Therefore, many men with increased PSA values undergo a biopsy sampling
3-5,
although this
RI PT
procedure is invasive, offers a low level of accuracy (i.e. only 30% detection rate at the first biopsy) and is prone to various complications, including sepsis and death 6, 7. Consequently, there is a need for more sensitive diagnostic method 3. Dogs are used for detecting explosives and drugs through their olfactory system, which can perceive thresholds as low as parts per trillion 8. As outlined by Pauling et
SC
al. in 1971, volatile organic compounds (VOCs) can be identified in human urine samples 9, 10. Several studies have shown that dogs may be trained to identify cancer patients by tracing the presence of a unique ‘‘odour signature’’. 11
provided the first evidence regarding “sniffer dogs’’. In 2001,
M AN U
In 1989, Williams and Pembroke
Church and Williams 12 reported a 66-year-old man who developed a patch of eczema at which a pet Labrador sniffed persistently. Histopathology demonstrated a basal cell carcinoma. Since 2001, several studies reported the detection of bladder 13-15, lung 16, skin 17, breast 16, and ovarian 18, 19 cancers and infectious diseases 20 using a dog’s sense of smell. Gordon et al. and Cornu et al. have extended the use of detection dogs to PC 21, 22. Although Gordon et al. did not produce positive results they did point out
TE D
procedural errors that needed to be addressed by later researchers
21.
They concluded:
"Unfortunately, the study did not yield success but offered valuable lessons in the form of mistakes, presented here with the hope other people will benefit from them" 21. Cornu et al. showed a sensitivity and specificity of 91% 22. Following on from Gordon et al. 21, Cornu et al. 22 took a step forward, though
EP
they themselves indicated important biases, including the limited series of patients enrolled, the use of only one dog and the control group which included patients aged >50 years with PSA > 8 ng/ml thus
AC C
with a high-risk of non-detected PC. The opinion that “the most problematic issue that has emerged so far is the large heterogeneity of performance across the different studies as well as within the same study”
10,
together with the limited cohort of patients and non-standardized training methodologies
has led us to design an accurate procedure to investigate whether dog olfactory detection remains a myth or can be a real clinical opportunity. Here, we assessed the diagnostic accuracy, in term of sensitivity and specificity at which a rigorously trained canine olfactory system can recognize PC-specific VOCs in urine from a large series of patients with PC at different stages and grades versus a heterogeneous control group.
MATERIALS AND METHODS
Participants
3
ACCEPTED MANUSCRIPT A total of 902 participants were recruited between November 2012 and November 2013 at the Humanitas Research Hospital. The participants were placed in two main Groups: a) Prostate Cancer Group and b) Control Group. Prostate Cancer Group (n = 362) a) Patients (n = 180) who have undergone either radical prostatectomy (open and robotic surgery).
RI PT
b) Men (n = 120) with elevated serum PSA levels (>2.5 ng/ml) or to abnormal Digital Rectal Examination (DRE). All patients had undergone prostate biopsy sampling only patients who presented a histological diagnosis of PC.
23.
We investigated
c) Patients (n = 22) with incidental PC detected during Transurethral Resection of the Prostate
SC
(TUR-P) or with very low- and low-risk PC on active surveillance.
d) Patients (n = 29) with “metastatic PC” or under hormonal therapy for biochemical relapse. e) Patients (n = 11) with synchronous primary PC and other different tumors.
M AN U
All these participants were consecutively recruited. For each subject a urine sample was spontaneously collected before the prostate biopsy sampling, radical prostatectomy or TUR-P.
Control Group (n = 540)
a) Non-pregnant, healthy, younger and older female volunteers (n = 50). b) Female patients (n = 72) affected by non-neoplastic (i.e. urinary infection, urolithiasis,
TE D
neurological or metabolic disorders, obesity, hyperthyroidism or hypertension), or with bladder, breast, kidney, ovary, vulva, uterus, stomach, colon, liver, skin, blood, tonsil or pancreas cancers.
c) Healthy young male volunteers (age: 18-25 years) with negative family history of PC (n = 60).
EP
d) Adult men (age > 45 years, n=240) with negative family history of PC, negative DRE and serum PSA levels < 1 ng / ml or < 2.5 ng / ml stable in time, with urological and/or systemic diseases.
AC C
e) Men with serum PSA levels < 2.5 ng / ml stable in time affected by urinary obstruction who had undergone TUR-P for benign prostatic hyperplasia (n = 40). f) Men (n = 78) with serum PSA levels < 2.5 ng / ml stable in time, with negative family history of PC, negative DRE and non-prostatic cancers.
Some participants were under pharmacologic treatments (Table 1S). No exclusion criteria were assumed with regard to a subject’s medical history, alcohol consumption, drugs, food, tobacco and other habits. Each participant was informed regarding the study and signed given consensus information. The Ethical Committee of the Center treating the patients (Humanitas Clinical and Research Center, Rozzano, Milan, Italy) approved the study (n. CE ICH - 260/11).
Urine samples
4
ACCEPTED MANUSCRIPT For each subject a urine sample (roughly, 30 ml) was spontaneously collected before the prostate biopsy sampling or radical prostatectomy in a sterile urine container and immediately stored at -20°C in different compartments according to the PC versus Control Group to avoid potential contaminations. These were then transported at controlled temperatures to the CEMIVET and stored at -20° C until used (Figure 1 and Supplementary Materials).
RI PT
Canine training process
A professional team trained two three-year old female German Shepherd Explosive Detection Dogs (EDD), (Zoe, chip n. 961100000401699 and Liu, chip n. 941000002579688). The team of the Italian Ministry of Defense’s, Military Veterinary Center (CEMIVET, Grosseto, Italy) was made up of 4 expert
SC
persons: a chief medical veterinary surgeon, a head trainer and 2 handlers. The dogs were trained using the “clicker training method” (i.e. operant conditioning). The dogs were taught to sit in front of the sample of urine recognized as cancer-infected (Figure 1, Video 1 and 2). Training was a full-time
M AN U
job for the team. During the training phase (from June 2012 to October 2012) 200 urines from the PC Group and 230 urines from the Control Group were analyzed and then not re-used in the “evaluation phase”.
Runs were blind for all team excepting the chief medical veterinary surgeon, who observed the runs from outside the set-up room. The team involved in the “training” and “evaluation phase” were not informed about the demographic and clinical characteristics of the participants. The purpose behind
TE D
our “training procedure” was to teach Zoe and Liu to recognize and store a pool of VOCs specifically present in the urine of patients with PC while ignoring urine free from PC-specific or other interfering VOCs. The various stages of the training set were undertaken in parallel with the two dogs. To define a standardized “working methodology”, the training program and materials used were designed
EP
specifically for the experimental procedure (Supplementary Materials). In brief, after being defrosted, 2 ml of the individual urine samples were housed in circular perforated metal containers to
AC C
allow the passage of odors (Figure 2). The metal containers were then placed in thermally sealed plastic packets, to avoid the operator-dependent or other contaminations. The “first phase” of the “training procedure” involved the use of urine from patients of the PC Group versus that from healthy young female participants. The choice of healthy female participants was dictated by the need to be absolutely certain that no specific prostate-VOCs could confuse the work of the two dogs. After registering the positive feedback of the two dogs we moved on to the “second phase”, which involved urine from the PC Group versus that from women of various ages with non- neoplastic or neoplastic diseases and undergoing various drug therapies. After registering the positive feedback of the two dogs we moved on to “third phase”, which compared urine from the PC Group with that from healthy young male volunteers. After registering the positive feedback of the two dogs we moved on to the “fourth phase”, which concentrated on urine from patients of the PC Group versus that from adult men. To gradually increase the complexity of the test and exclude conditioned reflexes in the dogs, the runs
5
ACCEPTED MANUSCRIPT were designed in the following way: 1 positive sample followed by 2 negative ones. The number of negative samples was then increased up to a ratio of 1 positive sample to 5 negative ones. We decided to evaluate single runs of 6 samples. The number of positive samples was also progressively increased: 0 out of 6, 1 out of 6, 2 out of 6, 3 out of 6, 4 out of 6, 5 out of 6 and 6 out of 6. We simultaneously ran checks containing no positive samples at all in order to test the dogs’ accuracy and exclude the possibility that the animals might report a not existent find simply to earn the reward. The positioning
RI PT
of the positive samples was always random determined by a ad hoc software. The dogs reported the positive sample only after making the full run twice. The results of the training procedure have been not considered for statistical purposes.
SC
Canine testing evaluation
After completing the “training phase” we proceeded to the “evaluation phase”. The run scheme used contained 6 urine samples. All urine samples from both the PC and Control Groups were blinded
M AN U
analyzed. Both dogs tested all samples (n = 902) after their random positioning. Concluded the daily runs, the Chief Medical Veterinary Surgeon checked the results and collected all the data. According to Cornu et al.
22
after each failure a new run was conducted three times using the same and other
random samples. The diagnostic performance has been evaluated excluding the re-runs.
Statistical analysis
TE D
Baseline demographic and clinical characteristics were summarized using descriptive statistics (median and min-max for continuous variables, and absolute frequencies for categorical variables). Diagnostic test performance was described using the sensitivity (i.e. the conditional probability of the dog’s indicating cancer where the condition existed), the specificity (i.e. the conditional probability of
EP
the dog’s ignoring a sample from a healthy donor) and the positive and negative likelihood ratios (i.e. LR+: the ratio of the conditional probability of the dog’s indicating cancer where the condition existed
AC C
to the conditional probability of the dog’s indicating cancer where the condition didn’t exist; LR-: the ratio of the conditional probability of the dog’s ignoring a sample from a sick donor to the conditional probability of the dog’s ignoring a sample from a healthy donor); for the sensitivity and specificity parameters the simple proportion as point estimate and the two-sided 95% Clopper-Pearson confidence limits were reported; for the diagnostic likelihood ratios the rate ratio as point estimate and the two-sided exact 95% confidence intervals were reported; diagnostic test performance was evaluated over the whole study population and excluding the female participants or considering only adult men in the control group. An exact logistic regression model was used to test whether baseline demographic and clinical characteristics were statistically associated to the dogs’ ability to recognize PC-specific VOCs; all statistical tests were two-sided and an exact p-value
PII: DOI: Reference:
S0022-5347(14)04573-X 10.1016/j.juro.2014.09.099 JURO 11844
To appear in: The Journal of Urology Accepted Date: 19 September 2014 Please cite this article as: Taverna G, Tidu L, Grizzi F, Torri V, Mandressi A, Sardella P, La Torre G, Cocciolone G, Seveso M, Giusti G, Hurle R, Santoro A, Graziotti P, Highly-Trained Dogs’ Olfactory System Detects Prostate Cancer in Urine Samples, The Journal of Urology® (2014), doi: 10.1016/ j.juro.2014.09.099. DISCLAIMER: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our subscribers we are providing this early version of the article. The paper will be copy edited and typeset, and proof will be reviewed before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to The Journal pertain.
Embargo Policy All article content is under embargo until uncorrected proof of the article becomes available online. We will provide journalists and editors with full-text copies of the articles in question prior to the embargo date so that stories can be adequately researched and written. The standard embargo time is 12:01 AM ET on that date. Questions regarding embargo should be directed to [email protected].
ACCEPTED MANUSCRIPT Highly-Trained Dogs’ Olfactory System Detects Prostate Cancer in Urine Samples
Gianluigi Taverna1, Lorenzo Tidu2, Fabio Grizzi1, Valter Torri3, Alberto Mandressi1, Paolo Sardella2, Giuseppe La Torre2, Giampiero Cocciolone2, Mauro Seveso1, Guido Giusti1, Rodolfo Hurle1, Armando
1Humanitas
RI PT
Santoro1, Pierpaolo Graziotti1
Clinical and Research Center, Rozzano, Milan, Italy; 2Italian Ministry of Defense’s, Military
Veterinary Center, CEMIVET, Grosseto, Italy; 3IRCCS Istituto di Ricerche Farmacologiche “Mario Negri”,
M AN U
SC
Milan, Italy.
Notes: This study was partially presented in the Podium Session at the Annual Meeting of the
TE D
American Urological Association, May 16-21, 2014.
Correspondence to: Gianluigi Taverna, Head of Prostate Diseases Section, Department of Urology, Humanitas Clinical and Research Center, Via Manzoni 56 - Rozzano – Milan, Italy Phone: +39 02
AC C
EP
82244535; Fax: +39 02 82244790; E-mail: [email protected]
1
ACCEPTED MANUSCRIPT Abstract Purpose: To establish the diagnostic accuracy, in term of sensitivity and specificity at which a rigorously trained canine olfactory system can recognize prostate cancer (PC)-specific volatile organic compounds (VOCs) in urine samples. Materials and Methods: Two three-year old female German Shepherd Explosive Detection Dogs were trained to identify PC-specific VOCs in urine samples and tested on 902 subjects (362 with PC ranging
RI PT
from very-low risk to metastatic and 540 healthy, affected by non-neoplastic diseases or non-prostatic tumors control participants). This cross sectional design for diagnostic accuracy involved one large Italian teaching hospital and the Italian Ministry of Defense’s, Military Veterinary Center.
Results: The dogs achieved the following performances: Dog 1: sensitivity 100% (95%CI: 99.0-
SC
100.0%) and specificity 98.7% (95%CI: 97.3-99.5%). Dog 2: sensitivity 98.6% (95%CI: 96.8-99.6%) and specificity 97.6% (95%CI: 95.9-98.7%). When only the adult men in the Control Group were considered, Dog 1 achieved a sensitivity of 100% and specificity of 98% (95%CI: 96-99.2%), while Dog
M AN U
2 a sensitivity of 98.6% (95%CI: 96.8-99.6%) and specificity of 96.4% (95%CI: 93.9-98.1%). Analysis of false positive cases did not reveal any consistent pattern in terms of participant demographics or tumor characteristics.
Conclusions: A trained canine olfactory system detects PC-specific VOCs in urine samples with high estimated sensitivity and specificity. Further studies will be necessary to investigate the potential
TE D
predictive value of using this procedure for recognizing PC.
Keywords: Prostate; Cancer; Dog Olfactory System; Volatile Organic Compounds; Sensitivity;
AC C
EP
Specificity.
2
ACCEPTED MANUSCRIPT Introduction Prostate cancer (PC) represents the fifth most frequent cancer in the world 1. Although the prostate specific antigen (PSA) blood testing has increased PC detection, the main drawback is its lack of specificity and accuracy. High serum PSA levels can be detected in men with non-malignant conditions 2.
Therefore, many men with increased PSA values undergo a biopsy sampling
3-5,
although this
RI PT
procedure is invasive, offers a low level of accuracy (i.e. only 30% detection rate at the first biopsy) and is prone to various complications, including sepsis and death 6, 7. Consequently, there is a need for more sensitive diagnostic method 3. Dogs are used for detecting explosives and drugs through their olfactory system, which can perceive thresholds as low as parts per trillion 8. As outlined by Pauling et
SC
al. in 1971, volatile organic compounds (VOCs) can be identified in human urine samples 9, 10. Several studies have shown that dogs may be trained to identify cancer patients by tracing the presence of a unique ‘‘odour signature’’. 11
provided the first evidence regarding “sniffer dogs’’. In 2001,
M AN U
In 1989, Williams and Pembroke
Church and Williams 12 reported a 66-year-old man who developed a patch of eczema at which a pet Labrador sniffed persistently. Histopathology demonstrated a basal cell carcinoma. Since 2001, several studies reported the detection of bladder 13-15, lung 16, skin 17, breast 16, and ovarian 18, 19 cancers and infectious diseases 20 using a dog’s sense of smell. Gordon et al. and Cornu et al. have extended the use of detection dogs to PC 21, 22. Although Gordon et al. did not produce positive results they did point out
TE D
procedural errors that needed to be addressed by later researchers
21.
They concluded:
"Unfortunately, the study did not yield success but offered valuable lessons in the form of mistakes, presented here with the hope other people will benefit from them" 21. Cornu et al. showed a sensitivity and specificity of 91% 22. Following on from Gordon et al. 21, Cornu et al. 22 took a step forward, though
EP
they themselves indicated important biases, including the limited series of patients enrolled, the use of only one dog and the control group which included patients aged >50 years with PSA > 8 ng/ml thus
AC C
with a high-risk of non-detected PC. The opinion that “the most problematic issue that has emerged so far is the large heterogeneity of performance across the different studies as well as within the same study”
10,
together with the limited cohort of patients and non-standardized training methodologies
has led us to design an accurate procedure to investigate whether dog olfactory detection remains a myth or can be a real clinical opportunity. Here, we assessed the diagnostic accuracy, in term of sensitivity and specificity at which a rigorously trained canine olfactory system can recognize PC-specific VOCs in urine from a large series of patients with PC at different stages and grades versus a heterogeneous control group.
MATERIALS AND METHODS
Participants
3
ACCEPTED MANUSCRIPT A total of 902 participants were recruited between November 2012 and November 2013 at the Humanitas Research Hospital. The participants were placed in two main Groups: a) Prostate Cancer Group and b) Control Group. Prostate Cancer Group (n = 362) a) Patients (n = 180) who have undergone either radical prostatectomy (open and robotic surgery).
RI PT
b) Men (n = 120) with elevated serum PSA levels (>2.5 ng/ml) or to abnormal Digital Rectal Examination (DRE). All patients had undergone prostate biopsy sampling only patients who presented a histological diagnosis of PC.
23.
We investigated
c) Patients (n = 22) with incidental PC detected during Transurethral Resection of the Prostate
SC
(TUR-P) or with very low- and low-risk PC on active surveillance.
d) Patients (n = 29) with “metastatic PC” or under hormonal therapy for biochemical relapse. e) Patients (n = 11) with synchronous primary PC and other different tumors.
M AN U
All these participants were consecutively recruited. For each subject a urine sample was spontaneously collected before the prostate biopsy sampling, radical prostatectomy or TUR-P.
Control Group (n = 540)
a) Non-pregnant, healthy, younger and older female volunteers (n = 50). b) Female patients (n = 72) affected by non-neoplastic (i.e. urinary infection, urolithiasis,
TE D
neurological or metabolic disorders, obesity, hyperthyroidism or hypertension), or with bladder, breast, kidney, ovary, vulva, uterus, stomach, colon, liver, skin, blood, tonsil or pancreas cancers.
c) Healthy young male volunteers (age: 18-25 years) with negative family history of PC (n = 60).
EP
d) Adult men (age > 45 years, n=240) with negative family history of PC, negative DRE and serum PSA levels < 1 ng / ml or < 2.5 ng / ml stable in time, with urological and/or systemic diseases.
AC C
e) Men with serum PSA levels < 2.5 ng / ml stable in time affected by urinary obstruction who had undergone TUR-P for benign prostatic hyperplasia (n = 40). f) Men (n = 78) with serum PSA levels < 2.5 ng / ml stable in time, with negative family history of PC, negative DRE and non-prostatic cancers.
Some participants were under pharmacologic treatments (Table 1S). No exclusion criteria were assumed with regard to a subject’s medical history, alcohol consumption, drugs, food, tobacco and other habits. Each participant was informed regarding the study and signed given consensus information. The Ethical Committee of the Center treating the patients (Humanitas Clinical and Research Center, Rozzano, Milan, Italy) approved the study (n. CE ICH - 260/11).
Urine samples
4
ACCEPTED MANUSCRIPT For each subject a urine sample (roughly, 30 ml) was spontaneously collected before the prostate biopsy sampling or radical prostatectomy in a sterile urine container and immediately stored at -20°C in different compartments according to the PC versus Control Group to avoid potential contaminations. These were then transported at controlled temperatures to the CEMIVET and stored at -20° C until used (Figure 1 and Supplementary Materials).
RI PT
Canine training process
A professional team trained two three-year old female German Shepherd Explosive Detection Dogs (EDD), (Zoe, chip n. 961100000401699 and Liu, chip n. 941000002579688). The team of the Italian Ministry of Defense’s, Military Veterinary Center (CEMIVET, Grosseto, Italy) was made up of 4 expert
SC
persons: a chief medical veterinary surgeon, a head trainer and 2 handlers. The dogs were trained using the “clicker training method” (i.e. operant conditioning). The dogs were taught to sit in front of the sample of urine recognized as cancer-infected (Figure 1, Video 1 and 2). Training was a full-time
M AN U
job for the team. During the training phase (from June 2012 to October 2012) 200 urines from the PC Group and 230 urines from the Control Group were analyzed and then not re-used in the “evaluation phase”.
Runs were blind for all team excepting the chief medical veterinary surgeon, who observed the runs from outside the set-up room. The team involved in the “training” and “evaluation phase” were not informed about the demographic and clinical characteristics of the participants. The purpose behind
TE D
our “training procedure” was to teach Zoe and Liu to recognize and store a pool of VOCs specifically present in the urine of patients with PC while ignoring urine free from PC-specific or other interfering VOCs. The various stages of the training set were undertaken in parallel with the two dogs. To define a standardized “working methodology”, the training program and materials used were designed
EP
specifically for the experimental procedure (Supplementary Materials). In brief, after being defrosted, 2 ml of the individual urine samples were housed in circular perforated metal containers to
AC C
allow the passage of odors (Figure 2). The metal containers were then placed in thermally sealed plastic packets, to avoid the operator-dependent or other contaminations. The “first phase” of the “training procedure” involved the use of urine from patients of the PC Group versus that from healthy young female participants. The choice of healthy female participants was dictated by the need to be absolutely certain that no specific prostate-VOCs could confuse the work of the two dogs. After registering the positive feedback of the two dogs we moved on to the “second phase”, which involved urine from the PC Group versus that from women of various ages with non- neoplastic or neoplastic diseases and undergoing various drug therapies. After registering the positive feedback of the two dogs we moved on to “third phase”, which compared urine from the PC Group with that from healthy young male volunteers. After registering the positive feedback of the two dogs we moved on to the “fourth phase”, which concentrated on urine from patients of the PC Group versus that from adult men. To gradually increase the complexity of the test and exclude conditioned reflexes in the dogs, the runs
5
ACCEPTED MANUSCRIPT were designed in the following way: 1 positive sample followed by 2 negative ones. The number of negative samples was then increased up to a ratio of 1 positive sample to 5 negative ones. We decided to evaluate single runs of 6 samples. The number of positive samples was also progressively increased: 0 out of 6, 1 out of 6, 2 out of 6, 3 out of 6, 4 out of 6, 5 out of 6 and 6 out of 6. We simultaneously ran checks containing no positive samples at all in order to test the dogs’ accuracy and exclude the possibility that the animals might report a not existent find simply to earn the reward. The positioning
RI PT
of the positive samples was always random determined by a ad hoc software. The dogs reported the positive sample only after making the full run twice. The results of the training procedure have been not considered for statistical purposes.
SC
Canine testing evaluation
After completing the “training phase” we proceeded to the “evaluation phase”. The run scheme used contained 6 urine samples. All urine samples from both the PC and Control Groups were blinded
M AN U
analyzed. Both dogs tested all samples (n = 902) after their random positioning. Concluded the daily runs, the Chief Medical Veterinary Surgeon checked the results and collected all the data. According to Cornu et al.
22
after each failure a new run was conducted three times using the same and other
random samples. The diagnostic performance has been evaluated excluding the re-runs.
Statistical analysis
TE D
Baseline demographic and clinical characteristics were summarized using descriptive statistics (median and min-max for continuous variables, and absolute frequencies for categorical variables). Diagnostic test performance was described using the sensitivity (i.e. the conditional probability of the dog’s indicating cancer where the condition existed), the specificity (i.e. the conditional probability of
EP
the dog’s ignoring a sample from a healthy donor) and the positive and negative likelihood ratios (i.e. LR+: the ratio of the conditional probability of the dog’s indicating cancer where the condition existed
AC C
to the conditional probability of the dog’s indicating cancer where the condition didn’t exist; LR-: the ratio of the conditional probability of the dog’s ignoring a sample from a sick donor to the conditional probability of the dog’s ignoring a sample from a healthy donor); for the sensitivity and specificity parameters the simple proportion as point estimate and the two-sided 95% Clopper-Pearson confidence limits were reported; for the diagnostic likelihood ratios the rate ratio as point estimate and the two-sided exact 95% confidence intervals were reported; diagnostic test performance was evaluated over the whole study population and excluding the female participants or considering only adult men in the control group. An exact logistic regression model was used to test whether baseline demographic and clinical characteristics were statistically associated to the dogs’ ability to recognize PC-specific VOCs; all statistical tests were two-sided and an exact p-value
