Dr. Piccirillo presents an interesting concept. Although the knowledge that comorbidity and severity of symptoms have a bearing on the prognosis of a patient with cancer is not new, the attempt to measure this influence and include it into a reproducible staging system is commendable.
Dr. Piccirillo's paper addresses one important deficiency in our ability to account for excess mortality due to conditions related to and accompanying a particular cancer, which are not included in the classic measurements and anatomic descriptions of cancer staging developed by the American Joint Commission on Cancer (AJCC) or the International Union Against Cancer [1,2]. As the author well states, the tumor, node, metastasis (TNM) staging system as we know it today has shortcomings. Staging of cancer is not a fixed science. As new information becomes available, the classification and staging of cancer will change. At present, the anatomic extent of cancer is the primary basis for staging. The editors of the AJCC manual have stated that, "in the future, biologic markers and other factors may play a part" in cancer staging .
Value of Comorbidity in Aiding Treatment Decisions Unclear
The deficiencies highlighted by Dr. Piccirillo complicate the clinical researcher's task in evaluating the effectiveness of treatment regimens. The addition of comorbidity to the TNM staging system is an important innovation, and Dr. Piccirillo's methodology and rigorous study represent important contributions to the field of oncology research. However, although it is clear that such a scoring system would be valuable in the retrospective assessment of groups of patients, its utility in helping clinicians make treatment decisions for the individual patient is certainly not as clear-cut. The need for validation through prospective studies is nondebatable.
Treatment decisions for individual patients are based not only on the presence of serious comorbid conditions but also on our ability to treat such problems. Combining all patients with prognostic comorbidity (into Dr. Piccirillo's so-called gamma category) and making treatment decisions based on this categorization may introduce a significant bias against appropriate cancer treatment, especially for patients with early-stage disease. Since a specific treatment increases mortality risk in patients with some comorbid conditions but not in others, all patients with comorbid disease are not alike with respect to mortality risk, and thus, probably should not be combined.
Furthermore, although some treatment decisions may hinge on the presence or absence of serious comorbid conditions, many more are characterized by the necessity to accommodate patients who are simply "too sick" for therapy, whether due to comorbid conditions or the underlying malignancy. As comorbid conditions present varying grades of severity and stages of progression, their influence is independently confounding as well. It is possible that the addition of a scoring system, such as the Goldman Assessment of Cardiovascular Risk or the American Society of Anesthesiology Coding System used for surgical patients, could refine the prospective assessment of individual patients sufficiently to permit categorization.
Consolidation of Many Factors May Prove Difficult
Conjunctive consolidation is an excellent method for combining stage and comorbidity in a way that may be clinically meaningful. It allows the investigator to form clinically meaningful groups of patients for whom survival probabilities may be estimated. In the case where only two factors, staging and comorbidity, are being combined and each factor has a small number of levels (four and two, respectively), the results can be readily described. Eight groups are formed, and the resulting probabilities are easily interpretable. More than three factors and multiple levels within each factor would yield a large number of groups to be consolidated. Since the author suggests that age, symptom severity, performance status, and comorbidity are important for prognostication, an exponential expansion of stage groupings may be inevitable. Logistic regression, while eschewed by the author, may be useful for determining the relative importance of factors to be investigated.
In this study, four stages were combined with the presence or absence of comorbidity to produce eight prognostic groups. These eight groups were then consolidated to form three prognostic groups. One group, patients with significant comorbidities, had the poorest survival, regardless of TNM staging. Among patients without significant comorbidities, staging was an important factor to consider. Patients in stages III and IV had poorer survival than patients in stages I and II. Calling the prognostic groups alpha, beta, and gamma hides the descriptive properties of the groups. From a clinical perspective, describing a patient as having stage I or II and no comorbidity is more meaningful than characterizing him or her as an alpha patient.
The author's conclusion that there are three prognostic groups is interesting. Does this mean that in patients with cancer of the larynx, comorbidities do not need to be staged? In the comorbidity group, stage is not a prognostic factor. As correctly pointed out by the author, prospective collection of data from a different cohort of patients is necessary to demonstrate the validity of the proposed staging system. Perhaps 5-year survival is not the only end point of interest.
It is not clear from the article how the author consolidated the eight groups into three prognostic groups other than by placing cryptic brackets in the appropriate places. The data could not have been more perfect for an eyeball analysis. If survival rates were always so clearly defined by prognostic groups, statisticians would have to find other sources of employment. It would be useful to see the steps the author used to consolidate this data set, as well as the data set for which survival rates overlapped.
Conjunctive Consolidation vs Regression Analysis
Conjunctive consolidation is a descriptive technique with analytical properties. The author shows, perhaps tediously, that a parsimonious description can be analyzed using chi-squares, proportionate reduction in variance, and other statistical scores and tests. These steps in the analysis are as cumbersome as those used in a regression analysis.
Logistic regression is an analytical technique with descriptive properties. It calculates the probability that a patient is in a specified group based on a mathematical combination of prognostic factors. Chi-square goodness-of-fit statistics, sensitivity and specificity of cutoff points, classification tables, and ROC (Receiver Operating Characteristic) curves are easily generated.
Conjunctive consolidation and logistic regression are two ways of looking at the same data set, each of which has advantages and disadvantages. If the mathematical techniques of cluster analysis were applied to conjunctive consolidation, the resulting equations might be just as cryptic as those of logistic regression. Just as has been stated regarding statistics, what the numbers show is interesting, but what they "omit or hide" may be vital. If one focuses on the mathematics of regression, one loses sight of the results. A logistic regression analysis of these data should lead to identical results if one were to look at the classification table that most statistical software provides. Additional studies comparing the two techniques may afford us with insights into how, where, and when each method is more useful, as well as provide information about prognostic factors in cancer patients.
Inclusion of Biologic Data in Staging Systems
We need to look for ways to include biologic data in our cancer staging system . The shortcomings of the present TNM staging system in head and neck cancer have been described, and modifications have been suggested [4,5]. As one example, in cancer of the base of the tongue and tonsillary fossa, stage III and stage IV "favorable and unfavorable" groups have been recognized. Also, cancer of the ethmoid sinus cannot be staged under the present TNM system; clearly there are early and advanced, "favorable and unfavorable" cancers at this site.
To cite yet another example, among nonsmoking, young adults, squamous cell carcinoma of the oral tongue is a disease with a special aggressive behavior in some patients; stage by stage, these cancers behave more aggressively than those seen in the older, smoking, "traditional" head and neck cancer population. Should this be considered a function of comorbidity or a molecular property of the cancer? We certainly need additional information and prospective studies utilizing reproducible biologic probes to answer some of these questions.
Today's health-care system demands outcome studies that use different end points than traditional measurements, such as "disease-specific" survival after cancer treatment. Information such as that obtained by Dr. Piccirillo could be incorporated into measurement of outome and then used for the purpose of making decisions regarding treatment and allocating resources for such. Nevertheless, the data obtained through a retrospective study should be interpreted with caution and need to be validated before the method can be recommended for system-wide use. Premature application without prospective validation could have disastrous consequences.