Series I
Between them, the dogs correctly indicated all 42 reference samples, giving a sensitivity of 100%. Lotti correctly indicated 41 of the 42 test samples and wrongly indicated 2 of the 210 controls, giving a sensitivity of 97% and a specificity of 99%. Hanna correctly indicated 41 of the 42 test samples and made no erroneous identifications among the controls, giving a sensitivity of 97% and a specificity of 100% (Table
1). The combined results for both dogs showed a sensitivity of 97% and a specificity of 99% (Table
2).
Lotti | Yes | 42 | 41 | 2 | 10 | 6 | 2 | 10 | 6 | 5 |
No | 0 | 1 | 208 | 0 | 4 | 48 | 0 | 4 | 45 |
Hanna | Yes | 42 | 41 | 0 | 10 | 8 | 3 | 10 | 10 | 3 |
No | 0 | 1 | 210 | 0 | 2 | 47 | 0 | 0 | 47 |
Table 2
Sensitivity and specificity measures in Series I (both dogs together)
Sensitivity | 0.976 | 0.908 | 0.995 |
Specificity | 0.995 | 0.980 | 0.999 |
For any particular test result, the probability that it will be: |
Positive | 0.166 | 0.135 | 0.202 |
Negative | 0.833 | 0.797 | 0.864 |
For any particular positive test result, the probability that it is: |
True positive | 0.976 | 0.908 | 0.995 |
False positive | 0.023 | 0.004 | 0.091 |
For any particular negative test result, the probability that it is: |
True negative | 0.995 | 0.980 | 0.999 |
False negative | 0.004 | 0.000 | 0.0190 |
Each dog missed one test sample—one from Group B and one from Group C; both patients died of cancer—but still had a generally high sensitivity and specificity. There should be no doubt in assuming that, in the majority of patients, the number of characteristic odor molecules would have been limited compared with the reference material, and that it was this that led to the failures to identify test samples. In contrast, the dogs identified all reference samples correctly.
It is likely that the completion of surgery and chemotherapy reduced the number of cancer cells in the patients’ bodies, thereby reducing the number of odor molecules in their blood. Moreover, it seems likely that there were large individual quantitative differences in characteristic odor molecules in different samples. There will have been a wide range in the number of molecules in the samples, but the dogs were unable to signal quantitative differences; all they could do was to signal a positive or a negative result. However, the results are interesting because there were a number of patients who had radical surgery and subsequently received five courses of chemotherapy. All patients had clinical CR before the sixth course, as evaluated by palpation under general anesthesia and in some cases completed with a CT scan, and all except two had normal CA-125 values.
Generally, doctors do not know how many patients will have residual cancer cells after complete clinical remission is declared, and it is unknown whether the final treatment after this will kill any remaining cells. In fact, our results suggest that almost all of the patients in our study had viable cancer cells, and the majority (n=29) of the 42 patients died of their disease. In Group A, two patients died of intercurrent disease, one of ovarian carcinoma between 4 and 5 years after the treatment was finished, and one died of probable cancer, although without a diagnosed relapse. Nine patients survived until the 5-year follow-up. One of the dead patients in Group A had stage I/A, two had stage I/C and the fourth had Stage III. Of the nine surviving patients, two had Stage II/A, four had Stage I/A, and three had Stage I/B.
In Group B, all patients died within 2 years. In this group one patient had stage I/A, one had I/C and the remaining patients had stage III–IV. In Group C, one patient survived for 4 years relapse-free, but was then lost to follow-up. The remaining 16 patients died between 16 months and 5 years after the end of treatment. Two patients had stage I/C, one had IIB and 12 patients had Stage III or IV. A patient who was lost to follow-up had stage III disease.
We do not have information on clinicopathological features such as stage, tumor grade, histology, age and menopausal status of the individual patients included in Series I. However, our previous studies [
5,
6] clearly showed that when dogs were trained to recognize the smell of ovarian cancer, those variables did not affect sensitivity. Furthermore, although the study shows detection of cancer odor to be a very good prognostic factor, depending on the size of the group and the material selected, comparison with other known prognostic factors cannot be done.
On the other hand, it is also possible that surviving patients had residual living cancer cells between their fifth and sixth courses of treatment, but far fewer cells than patients who died of their disease, and the odor molecules from those cells were detected by the dogs.
It is difficult to discuss our results on a broader basis, because to our knowledge there are no other published studies that have used blood samples with trained dogs and related the findings to survival. However, our results strongly suggest a great need for a more sensitive marker than is currently available to ensure the safety of patients and increase the overall survival of ovarian cancer patients.
Six-month test samples
Lotti correctly indicated six of 10 test samples (one of which was uncertain) and wrongly indicated five of 50 controls, giving a sensitivity of 60% and a specificity of 90%. Hanna correctly indicated all 10 test samples (although four of the 10 selections were uncertain), and wrongly indicated three of 50 controls, giving a sensitivity of 100% and a specificity of 94% (Table
1). The combined results for both dogs showed a sensitivity of 80% and a specificity of 92% (Table
4).
Table 4
Sensitivity and specificity measures in Series II, 6 months (both dogs together)
Sensitivity | 0.8 | 0.557314 | 0.933894 |
Specificity | 0.92 | 0.843855 | 0.962321 |
For any particular test result, the probability that it will be: |
Positive | 0.2 | 0.134742 | 0.284927 |
Negative | 0.8 | 0.715073 | 0.865258 |
For any particular positive test result, the probability that it is: |
True positive | 0.666667 | 0.446926 | 0.835734 |
False positive | 0.333333 | 0.164266 | 0.553074 |
For any particular negative test result, the probability that it is: |
True negative | 0.958333 | 0.890714 | 0.986569 |
False negative | 0.041667 | 0.013431 | 0.109286 |
During the 2-day test for Series II, both dogs appeared to be under stress and generally unsettled; they barked, made various unusual sounds, and often turned back and tried to go in the opposite direction between the boxes. We suspected that this could have been due to an unusually low concentration of odor molecules in the test samples. We tested the dogs in the interval between the 2 test days without a test sample in the arrangement, and they performed the search as usual. The following day, when the test samples were included again, the anxious and insecure behavior returned. However, the fact that identification of reference samples was 100% correct and the specificity was high (as it was in our previous studies) suggests that the overall results are correct.
We selected samples from patients who were relapse-free during the first 3 years after treatment. This choice of time period was based on the fact that most recurrences are diagnosed in the first 3 years [
11]. Patients were selected from our database and since the disease may relapse even after the first 3 years after treatment, we also checked the patient case records in January 2013 (Table
5). The 3- and 6-month test samples of two patients (No. 8 and 9), were both clearly indicated by both dogs. For one additional patient (No. 3), the 3-month sample was indicated by both dogs while the 6-month sample was clearly indicated by Hanna and uncertainly indicated by Lotti. All three patients had recurrences, and two of them died 3–4 years after the end of treatment. The January 2013 record check showed that two of the 10 patients (No. 8 and 9) included in Series II who were thought to be recurrence-free during the first 3 years after treatment, had in fact relapsed a few months before 3 years had passed. The reason they were categorized as having no relapse was that the database had not been updated at the time of patient selection, and their current recurrence was not known to us.
Table 5
Tumor characteristics, dog responses, and survival for patients in Series II
1 | 4842/5095 | Feb 2009 | III/C | Seropapillary | 2 | O O | X # | Alive/relapse-free |
2 | 3932/4799 | June 2008 | III/C | Seropapillary | 3 | O X | X X | Alive/relapse-free |
3 | 406/1018 | Nov 2005 | III/C | Adenocarcinoma | 3 | X X | # X | Relapse 2008/Died 2009 |
4 | 9807/108 | Jan 2005 | II/C | Adenocarcinoma | 2 | O X | O # | Alive/relapse-free |
5 | 8647/8955 | Dec 2003 | III/C | Clear cell | 2 | X X | O X | alive/relapse-free |
6 | 10704/1328 | Dec 2005 | II/C | Seropapillary | 3 | # O | O # | Alive/relapse-free |
7 | 451/1065 | Sept 2005 | III/C | Undifferentiated | 4 | X X | O # | Relapse 2009/alive |
8 | 3835/6750 | Feb 2000 | III/C | Seropapillary | 3 | X X | X X | Relapse 2002/died 2005 |
9 | 6132/7872 | Jan 2004 | II/C | Seropapillary | 2 | X X | X X | Relapse 2006/alive |
10 | 4926/5214 | Aug 2008 | II/B | Seropapillary | 2 | O # | X X | Alive/relapse-free |
Six-month test samples from two patients (No. 2 and 10) were clearly indicated by the two dogs, but the patients remained relapse-free 4 years after completion of treatment. The dogs’ indications in those cases were fairly consistent, which may imply an increased risk of recurrence in future years.
The remaining indications were consistent with patient survival, although in some cases (e.g., No. 1 and 4) there were suggestions by uncertain indication behavior that there may be a few viable cancer cells remaining in the body. It is likely that in several cases the concentration of typical odor molecules was near the lower limit of canine detection ability. To estimate what this limit might be, we have previously published results showing that one dog (Hanna) was repeatedly able to identify with certainty a piece of fatty abdominal wall containing about 20 microscopically-verified ovarian cancer cells [
6]. It is impressive how this very low limit of detection allows dogs to signal probable future recurrences that would not be diagnosed by other methods for another 2–3 years. This is the most important result of the present study.
The dogs were able to indicate small numbers of living cancer cells with high sensitivity and specificity in a large group of ovarian cancer patients. To our knowledge; this is the first study to highlight the importance of characteristic odor molecules in the blood of ovarian cancer patients as a prognostic marker. Previously, McCulloch et al. described one patient with breast cancer in remission who was identified by dogs as having cancer [
2]. Detection of odor in the blood, currently only possible with trained dogs, can allow for early and long-term prediction of survival. An early diagnosis of primary or recurrent disease may also significantly improve the patient’s survival.