I Can't Believe He's Still Alive! Why Some Patients Don't Read the Books
ACVIM 2008
Dennis Bailey, DVM, DACVIM (Oncology)
Ithaca, NY, USA


We all have had cancer patients whose outcome far exceeded our expectations. There is no better feeling than the excitement felt when a long-time cancer survivor returns for a recheck appointment, full of energy and enjoying life to its fullest. But why does this happen? Sometimes it is because an individual's cancer truly has a different biologic behavior than what is expected. However, other times it is because we set our expectations too low. The published literature on veterinary oncology often is incomplete, forcing us to make assumptions that might not always be correct. In addition, the information present frequently is misinterpreted. To provide owners with accurate expectations, accurate interpretation of the current veterinary literature is critical. Below are several important things to consider when reading the primary literature. These items will be discussed in the context of several clinical cases.

Study Design

When Was the Study Performed?

It is important to know when the patients in a given study were treated to put the results in correct historical perspective. The biologic behavior of a given disease can change over time. For example, the decreased prevalence of feline leukemia virus (FeLV) infections has had a profound impact on feline lymphoma. In the 1970s and 1980s it was estimated that as many as 50-70% of all cats with lymphoma tested positive for FeLV antigenemia.1,2 Commercially-available FeLV diagnostic assays and FeLV vaccines were introduced in 1980s, and widespread integration of "test, eliminate, and protect programs" greatly reduced the prevalence of FeLV infection. Studies performed in 1990s and 2000s report that only 10-15% of cats with lymphoma tested positive for FeLV antigenemia.3-5 Associated with the epidemiologic change is a shift in the relative incidences of the different anatomic forms of lymphoma. The relative proportion of cranial mediastinal lymphomas has decreased from 40% to only 5-10%, while the relative proportion of gastrointestinal lymphomas has increased from 25% to 40-50%.1,3-6 Additionally, FeLV antigenemia is a strong negative prognostic factor that is associated with shorter remission durations and survival times.1,3,5

Even if the biologic behavior of the disease in question has not changed, the continual advances in veterinary medicine can seemingly improve the outcome of a given treatment. For example, while thoracic radiography historically has been the imaging modality of choice for detecting pulmonary metastasis, thoracic computed tomography (CT) is becoming more common. With CT, the threshold for reliable detection of pulmonary nodules is 1mm, compared to 7-9mm with radiography.7 Therefore, thoracic CT will identify more patients with early pulmonary metastasis. The reassignment of these patients to a higher clinical stage, called "stage migration," has the net effect of reducing the average tumor burden of the patients in both the lower and higher stage groups. The prognosis associated with aggressive local therapy improves because metastatic disease has been ruled out more accurately. Additionally, the prognosis associated with systemic chemotherapy improves because it is started earlier while tumor burden is lower and the patient hopefully is still asymptomatic.

Advances in supportive care also can improve the outcome associated with a given treatment. Early studies evaluating adrenalectomy for dogs with pheochromocytoma reported perioperative mortality rates around 50%.8 Perioperative cardiovascular complications were common secondary to excessive catecholamine release prior to and during tumor removal as well as to the rapid decline in catecholamine levels that occurred as soon as the tumor was removed. It is currently recommended that dogs be treated with phenoxybenzamine, a noncompetitive alpha-adrenergic antagonist, for 1-2 weeks pre-operatively. Additionally, if tachycardia is a problem, a beta blocker such as atenolol is recommended after the alpha blockade has been initiated. Preoperative adrenergic blockade, along with improvements in anesthetic protocols and patient monitoring, has reduced perioperative mortality rates to less than 20%.9

Defining the Treatment Population

The patient eligibility requirements for inclusion into the study should be stated clearly in the Materials and Methods. When evaluating these requirements, two questions should be asked. Would the patient I am treating have been eligible for this study? Would all dogs with a different cancer stage or different cancer altogether have been excluded from this study? If the answer to both of these questions is "yes," then the results of the study should provide a reasonable expectation regarding prognosis. However, if the answer to either of these questions is "no," then the results need to be interpreted cautiously. Sometimes the study in question clearly is irrelevant to the patient at hand. For example, studies evaluating adjuvant chemotherapy for dogs with osteosarcoma that do not have gross metastasis are not applicable to dogs with radiographic evidence of pulmonary metastasis. Determining the relevance of the patient population is not always this straight forward, though. For example, suppose we wanted to know the prognosis for a dog with subcutaneous hemangiosarcoma that staged negatively for metastasis and is treated with surgical excision of the primary tumor followed by adjuvant single-agent doxorubicin chemotherapy. Two studies that have evaluated outcome when dogs with hemangiosarcoma undergo surgical excision of their primary tumor followed by adjuvant doxorubicin chemotherapy, and they report median survival times of 6-8 months.10,11 However, both studies included dogs with primary tumors arising from a variety of locations. The vast majority of the dogs had primary splenic tumors, and only 20% had primary subcutaneous tumors. Based on studies evaluating prognosis with surgery alone, primary splenic tumors are much more aggressive than subcutaneous ones. When dogs with primary splenic tumors were treated with splenectomy alone, 85% of dogs developed gross metastasis and median survival was 3 months.12 In contrast, when dogs with primary subcutaneous hemangiosarcoma were treated with surgery alone, only 25% developed metastasis and median survival was 6-10 months.13 Based on this information, it is inappropriate to combine these two anatomic locations when assessing prognosis. The true median survival time for dogs with just subcutaneous hemangiosarcoma treated with surgery and doxorubicin likely is longer than the reported 6-8 months.

Prospective versus Retrospective Studies

Whenever possible, prospective studies are preferable over retrospective ones. In a prospective study, the entry criteria, staging diagnostics, treatment plan, and monitoring plan are established first, and then patients are enrolled into the study. Designing the study in advance helps improve its uniformity and minimizes the impact of any potentially confounding variables. Additionally, it is more likely that the data will be complete and all information of interest will be collected from every patient enrolled. For these reasons, the results of prospective studies are considered more reliable.

In contrast, a retrospective study draws its patient population from historical data. That is, all patients were diagnosed, staged, treated, and monitored before the study was initiated. Because the patients were not treated with the study in mind, there likely will be more variation with respect to the staging diagnostics, treatment plans, and monitoring plans. The potential for confounding increases greatly. For example, a recent retrospective multi-institutional study evaluated dogs with anal sac apocrine gland carcinomas.14 The dogs in this study were treated using several different treatment protocols: surgery alone, radiation therapy alone, chemotherapy alone, surgery and radiation therapy, surgery and chemotherapy, or surgery and radiation therapy and chemotherapy. Dogs treated with chemotherapy alone had significantly shorter survival times. At first glance, this might lead one to believe that chemotherapy is not an effective treatment for this form of cancer. However, dogs were not randomly assigned to treatment groups, and those receiving chemotherapy were more likely to have metastatic disease. Therefore, the poorer prognosis might be a reflection of more advanced disease rather than relative inefficacy of treatment.

Another limitation of retrospective studies is that the data is only as complete as the medical record. It usually is difficult or impossible to fill in missing data. In particular, it can be extremely challenging to retrospectively assess things like quality of life and adverse effects of treatment. There are well-defined criteria for grading the severity of adverse events,15 but grading can only be as accurate as the information written down in the medical record.

This is not to say that retrospective studies do not have value. They are particularly useful when evaluating rare tumors that are diagnosed so infrequently that it is not possible to prospectively accrue cases in a timely manner.

Defining a Control Group

The best way to determine the relative benefit of a given therapy is to randomly assign patients to a treatment group (that receives the given therapy) or a control group (that does not receive the given therapy) and then compare the outcome for each group. While the patients in the control group must not receive the therapy in question, that does not mean they should be refused treatment. When evaluating a new therapy, the ultimate goal is to determine whether or not this new therapy should replace the current standard of care. Therefore, the individuals in the control group should receive what is believed to be the current standard of care.

Difficulty arises, however, when there is no standard therapy, a situation all too common in veterinary oncology. Going back to the treatment of anal sac apocrine gland carcinoma in dogs, at this time there is no clear evidence supporting or refuting the benefit of chemotherapy. Additionally, when chemotherapy is recommended there is no uniformly accepted front-line protocol. In the previously mentioned retrospective study, the dogs treated with chemotherapy received cisplatin, carboplatin, doxorubicin, mitoxantrone, melphalan, actinomycin D, piroxicam, epirubicin, chlorambucil, or a combination of these drugs.14 So where does one begin when trying to figure out the merits of chemotherapy for the treatment of this cancer? The simplest thing to do would be to perform a prospective study where dogs are randomized to receive either local therapy alone or local therapy along with a uniform chemotherapy protocol. Several factors have prevented this from being done, though. First, while the benefit of chemotherapy has not been definitively demonstrated, most oncologists (myself included) still recommend its use based on the high metastatic rate associated with anal sac carcinomas and anecdotal experience. Consequently, while the benefit of chemotherapy has never been proven, withholding it from the dogs in the control group could be considered unethical. Second, the cost differential between the control group and the treatment group would be significant and difficult to reconcile without outside funding.

Historical controls often are used in the veterinary literature. They avoid some of the real or perceived ethical dilemmas associated with randomized controlled studies. However, their utility is limited by all of the factors discussed above (see "When was the study performed?"), and results using historical controls must be interpreted very cautiously.

Survival Statistics

When assessing the outcome of a given treatment protocol, there are several different endpoints to consider. Survival time usually receives the most focus, since this is the primary concern of pet owners. However, it might not be the most accurate reflection of treatment efficacy. If the treatment fails, owners might elect to pursue additional rescue therapy. Additionally, different owners have different opinions regarding quality of life and euthanasia. Because of this, many regard disease-free interval to be a better endpoint. Another endpoint that is gaining popularity is the "time to first event"--the time until disease recurrence or death, whichever occurs first.

Special statistical tests are used when evaluating disease-free interval and survival because frequently some of the patients enrolled in the study have not yet reached the endpoint in question when the data is analyzed. Consider, for example, a dog with lymphoma that is still alive and in a complete remission 5 years after starting therapy. It is unknown when, if ever, the lymphoma will recur. (Yes, on rare occasion dogs with lymphoma are cured!) If we analyze the data while the patient is alive, though, what should be done? The patient should not be eliminated from the study--considerable information has been gained from following him over the past 5 years. At the same time, it would be inappropriate to say that he is cured--unfortunately the lymphoma still might recur at any time. The most appropriate thing to do is to censor him at 5 years; that is, to acknowledge that we know he is alive and in remission today (5 years after starting therapy) but don't know what is going to happen beyond today.

Censoring is an important part of survival statistics, and every study should clearly define its censoring criteria. For example, when assessing survival some studies regard all deaths as endpoints irrespective of cause of death, whereas others only regard those deaths directly attributable to the cancer or treatment as endpoints. In the latter situation, dogs that die from unrelated causes are censored at their time of death. This practice is controversial because it can be difficult or impossible to be absolutely certain that a death was completely unrelated to the cancer or the treatment, and inappropriate censoring can falsely improve the calculated survival function.

Kaplan-Meier Curves

The Kaplan-Meier product limit method is the statistical method used most commonly when analyzing disease-free interval and survival. The details of how to calculate survival functions are beyond the scope of this lecture. However, the curves generated using this statistical method are found throughout the veterinary oncology literature, and it is important to be able to interpret them. Figure 1 is provided below as a reference.

Figure 1.
Figure 1.

Kaplan-Meier curve illustrating the disease-free interval for 24 dogs with appendicular osteosarcoma that underwent amputation or limb-sparing surgery followed by adjuvant carboplatin and doxorubicin chemotherapy.
(Modified from Bailey D, et al.16). Censored patients are indicated by vertical ticks.

Median disease-free interval (DFI) is defined as the time at which exactly 50% of the patients are disease-free. In Figure 1, median DFI is 195 days. Medians are used more commonly than means (averages) because they are less likely to be affected by outliers. (Consider, for example, a situation where 49 dogs remain disease-free for only 7 days and 1 remains disease free for 10 years. Median DFI would be 7 days, whereas mean DFI would be 80 days.).

Significant emphasis is placed on median DFI and survival, but the overall shape of the Kaplan-Meier curve is equally important. Does the curve begin to drop immediately? In Figure 1, the first drop occurs at 14 days, indicating that disease recurrence has the potential to occur very early. Does the curve ever level off? In Figure 1, the curve levels off around 300 days, indicating that this might be an important milestone where prognosis begins to improve. Where are the censored points? As in Figure 1, there should be relatively few censored points and they should occur predominantly in the later segments of the curve.

Limitations of Statistics

When counseling owners, it is important to ensure that they recognize both the value and limitations of cancer statistics.

Statistics are very helpful for providing a reasonable expectation of the most likely outcome when large numbers of patients with the same disease undergo the same treatment. They do not make any predictions about specific individuals. Any given individual might have an outcome that is far better or worse than our expectation. As with any aspect of veterinary medicine, there never are any guarantees. Statistics and estimates of prognosis therefore should be used only as a reference point when trying to decide whether to commit to the emotional and financial investment associated with treating cancer.


1.  Mooney SC, et al. J Am Vet Med Assoc 1989;194:696.

2.  Cotter SM. J Am Anim Hosp Assoc 1983;19:166.

3.  Milner RJ, et al. J Am Vet Med Assoc 2005;227:1118.

4.  Louwerens M, et al. J Vet Intern Med 2005;19:329.

5.  Vail DM, et al. J Vet Intern Med 1998;12:349.

6.  Mooney SC, et al. Semin Vet Med Surg (Small Anim) 1986;1:51.

7.  Nemanic S, et al. J Vet Intern Med 2006;20:508.

8.  Barthez PY, et al. J Vet Intern Med 1997;11:272.

9.  Kyles AE, et al. J Am Vet Med Assoc 2003;223:654.

10. Ogilvie GK, et al. J Vet Intern Med 1996;10:379.

11. Sorenmo KU, et al. J Vet Intern Med 2004;18:209.

12. Wood CA, et al. J Am Anim Hosp Assoc 1998;34:417.

13. Ward H, et al. J Vet Intern Med 1994;8:345.

14. Williams LE, et al. J Am Vet Med Assoc 2003;223:825.

15. Veterinary Co-operative Oncology Group. Vet Comp Oncol 2004;2:194.

16. Bailey D, et al. J Vet Intern Med 2003;17:199.

Speaker Information
(click the speaker's name to view other papers and abstracts submitted by this speaker)

Dennis Bailey, Jr., DVM, DACVIM (Oncology)
Ithaca, NY