Inclusion of Comorbidity in a Staging System for Head and Neck Cancer

September 1, 1995

The widespread use of the TNM staging system has helped standardize the classification of cancers. Despite its excellence in describing a tumor's size and extent of anatomic spread, the TNM system does not account for the clinical biology of the cancer.

The widespread use of the TNM staging system has helped standardize the classification of cancers. Despite its excellence in describing a tumor's size and extent of anatomic spread, the TNM system does not account for the clinical biology of the cancer. Clinical factors, such as symptom severity, performance status, and comorbidity, which are important for classification, prognostication, and evaluation of treatment effectiveness, remain excluded from this system. In several studies of cancer prognosis, the presence of severe comorbidity was found to dramatically influence survival statistics and the evaluation of treatment effectiveness. A statistical technique known as conjunctive consolidation was used to incorporate comorbidity into the TNM staging system and maintain the four category system. Utilizing this technique, comorbidity was added to the TNM system for laryngeal cancer to create a composite staging system. Quantitative evaluation of the new system showed that the addition of comorbidity provides improved prognostic precision over TNM stage alone.

Introduction

The tumor, node, metastasis (TNM) system of cancer classification was originally described by Denoix and Schwartz [1] in 1948 and, in 1953, was incorporated by the International Union Against Cancer (Union International Contre le Cancer, UICC) and the International Congress of Radiology into a formal classification system for cancer [2]. The TNM system was adopted in the US in 1959, when the American College of Surgeons, American College of Radiology, College of American Pathologists, American College of Physicians, American Cancer Society, and National Cancer Institute cosponsored the formation of the American Joint Committee for Cancer Staging and End Results Reporting [3]. Its stated purpose was "to develop systems for the clinical classification of cancer which would be of value to practicing American physicians" [4].

During the past 30 years, the AJCC has worked to develop TNM definitions and stage classifications for all disease sites and subsites, revise definitions and classifications based on the results of clinical studies, and educate physicians and other health-care professionals about the TNM classification system. In 1988, the AJCC reached agreement with the UICC for a common TNM and stage classification system, thereby resolving many minor differences that had previously prevented the creation of a single tumor classification system.

TNM-A Morphologic Staging System

The TNM system classifies a cancer's spread from primary to distant sites. In the general TNM approach, the six T categories-T0, Tis, T1, T2, T3, and T4-refer to the extent of the primary tumor; the four N categories-N0, N1, N2, and N3-denote the degree of involvement of regional nodes; and the two M categories-M0 and M1-designate the absence or presence of distant metastasis.

For each patient, the individual T, N, and M category ratings are combined in tandem to form expressions such as T2, N1, M0 or T3, N2, M1. Because six categories of T, four categories of N, and two categories of M create 48 possible ratings for the TNM expressions, stage groupings (I, II, III, and IV) were created by combining categories to ease statistical analyses. The TNM system is thus an exclusively morphologic classification, offering a reasonably precise description of the anatomic extent of tumor at a specific time.

Clinical Biology of Cancer

The form of the tumor can be expressed in gross anatomic terms (TNM index), microscopic terms (eg, cell type, degree of differentiation), and biomolecular terms (eg, tumor markers, ploidy). The function of the tumor can be described by clinical effects that create severity of illness within the patient. The tumor's functional effects are manifested by the type, duration, and severity of cancer symptoms (eg, weight loss, fatigue) [5-10] and by the performance or functional status of the host [11,12].

Another important aspect of clinical biology is the comorbidity, or the "setting" in which the cancer occurs. Although unrelated to the cancer itself, the patient's concomitant disease(s) can affect the clinical course of cancer, choice of treatment, and prognosis [13-17].

Shortcomings of TNM System

Despite its excellence in describing a tumor's size and extent of anatomic spread, the TNM system does not account for the clinical biology of the cancer [18-20]. A substantial amount of clinical research has demonstrated the prognostic shortcomings of the TNM system (unpublished data, JF Piccirillo, MD, and AR Feinstein, MD; and references 5, 12, and 21-24), while several editorials have requested improvements in it (unpublished data, JF Piccirillo, MD, and AR Feinstein, MD; and references 20, 25, and 26). Nevertheless, no significant modifications have been made in the TNM system since its creation. In particular, important patient-based prognostic factors, such as symptom type and severity, severity of comorbidity, and functional capacity, are excluded. Despite unequivocal evidence of the prognostic importance of symptoms, performance status, and comorbidity, the AJCC system remains relentlessly confined to morphologic data.

Prognostic Importance of Comorbidity

Many people with cancer also have other nonneoplastic diseases. These comorbidities may be so severe as to prohibit the use of preferred antineoplastic therapie and impact on 3- and 5-year survival statistics. In several studies of cancer prognosis, the presence of comorbidity was found to dramatically affect survival and the evaluation of treatment effectiveness [7,16,21,27-30].

Prognostic Comorbidity

Prognostic comorbidity [31] refers to comorbidity severe enough to impact on survival rates. Examples of prognostic comorbidity are significant cardiac disease, severe hypertension, far-advanced tuberculosis, severe liver disease, and recent severe stroke. Examples of nonprognostic comorbidity include a history of "mild" hypertension that is well controlled with medication, congestive heart failure or myocardial infarction of more than 6 months duration, recurrent asthma attacks without underlying lung disease, and slight gastrointestinal bleeding not requiring transfusion. As shown in Table 1, 5-year survival rates in a variety of cancers are dramatically worse in patients with prognostic comorbidity than in those without prognostic comorbidity.

Comorbidity vs Tumor Size and Stage

In many cancers, comorbidity is prognostically more important than tumor size or stage. The cancers for which comorbidity is particularly important are those which are not rapidly fatal and which affect people who are middle-aged or older (ie, over age 50 years). These include cancers of the breast, prostate,7 oral cavity, pharynx and larynx [16,29], bladder, ovary, and uterus [17], as well as non-Hodgkin's lymphoma. Based on recent incidence rates, these cancers represent approximately 61% of all cancers for men and 65% for women.

The importance of comorbidity is clear from these statistics, and yet the present cancer staging system does not contain this important information. As several valid comorbidity instruments now exist, the continued exclusion of standardized comorbidity information from cancer statistics appears to be a major omission.

Specific Comorbidity Instruments

Comorbidity instruments can be classified into two groups, depending on the origin of the data: (1) instruments that rely on primary data and (2) instruments that are based on secondary data. Primary data are collected from physicians or nurses or through chart reviews. Secondary data are derived from administrative and financial databases main- tained by hospitals, insurance companies, and state and federal governments.

Instruments Derived from Primary Data

Comorbidity measures that rely on primary data include the Kaplan-Feinstein Index [31], the Charlson Co-Morbidity Index [15], and the Index of Co-Existent Disease [32].

The Kaplan-Feinstein Comorbidity Index was developed from a study of the impact of comorbidity on 5-year survival outcomes for patients with diabetes mellitus. In this index, specific diseases and conditions are classified as mild, moderate, or severe based on the severity of organ decompensation. An overall comorbidity score is assigned according to the highest level of decompensation.

The Charlson Comorbidity Index was created from studies of 1-year mortality among patients admitted to a medical unit of a teaching hospital. It is a weighted index that takes into account the number and seriousness of comorbid diseases. The scoring system for this instrument assigns weights of 1, 2, 3, and 6 for each of the comorbid diseases present at initial assessment, and derives from these a total score that determines the patient's overall prognostic status.

The Index of Co-Existent Disease (ICED) predicts length of stay and resource utilization after hospitalization for surgical procedures. To calculate the overall burden of comorbidity, ICED assesses the patient's status in two separate components: physiologic and functional burden.

Impact of Comorbidity on Therapy

Aside from direct effects on survival, severe comorbidity can also have a prognostic impact by altering therapy. A patient who is "too sick" to tolerate a preferred treatment may be given a less aggressive or even palliative therapy. Consequently, the presence of severe comorbidity, rather than the TNM stage, may sometimes determine the selection of treatment [33-35] and the patient's eventual outcome [7,17,36]. Nevertheless, comorbidity data are not currently collected or included in cancer statistics. As demonstrated in Table 1 and the previously cited reports, this omission continues to produce major imprecision in the classification of patients and the subsequent interpretation of both 5-year survival rates and therapeutic effectiveness.

Inclusion of Comorbidity in a Staging System

We assessed the prognostic impact of comorbidity in a group of patients with cancer of the larynx and used a statistical techniques called conjunctive consolidation to include comorbidity in a staging system. The illustrative data derived from a study of patients with biopsy-proven squamous cell carcinomas of the larynx first treated between 1975 and 1983 at Yale-New Haven Hospital, New Haven, Connecticut [16]. The methods of research included retrospective review of the medical records and use of a standard medical record data extraction form, coding form, and coding criteria handbook.

Information was obtained in the following categories: standard demographic; symptom type, duration, and severity; medical comorbidities; tumor description; treatment; and 5-year follow-up. The severity of comorbidity was classified according to the Kaplan-Feinstein Index [31]. Patients were staged morphologically according to the 1988 AJCC TNM criteria [3]. The data were entered into a microcomputer for analysis with the SAS statistical system.

Baseline Classification and 5-Year Survival Data

The cohort of 196 patients with laryngeal cancers consisted of 151 men and 42 women whose median age was 62 years. There were 168 Caucasians, 24 African-Americans, 3 Hispanics, and 1 Asian-American. Table 2 shows the 5-year survival results according to eight baseline variables for the 193 patients for whom 5-year survival data were available. Age, symptom status, TNM stage, and prognostic comorbidity were the variables that provided important prognostic information.

Conjunctive Consolidation

To incorporate comorbidity into the TNM staging system and maintain the four category system, a statistical technique known as conjunctive consolidation, or targeted-cluster analysis, was used. This form of multivariate analysis can be used to incorporate prognostic variables into an existing staging system without relying on cryptic mathematical regression equations or the exponential expansion of stage groupings [26,37-39].

In conjunctive consolidation, prognostic variables are combined, based on biologic and statistical criteria, to produce composite variables. The system can maintain a discrete number of stages (usually three or four), and is more relevant and meaningful to clinicians than are the results of regression equations [39].

Prognostic Information Provided by Comorbidity-Utilizing the conjunctive consolidation techniques, we demonstrated the unique prognostic information provided by comorbidity within TNM. Table 3 shows the conjoined effects of comorbidity and TNM stages. For the 77 patients in TNM stage I, the survival rate is 78%. Within that stage group, however, the 71 patients without prognostic comorbidity have a survival rate of 83%, whereas the 6 patients with prognostic comorbidity have a rate of 17%. Likewise, within TNM stages II, III, and IV, the presence of prognostic comorbidity defines unique prognostic subgroups that would not have been identified by TNM stage alone.

Three Composite Stages-Because both prognostic comorbidity and TNM stages were distinctively important, we combined the categories, using the conjunctive consolidation strategy. Thus, the eight comorbidity-TNM stage groups were consolidated into three composite stages: alpha, beta, and gamma (Table 4).

Patients with prognostic comorbidity, regardless of TNM stage, formed the gamma group. Among the 166 patients without prognostic comorbidity, the beta group consisted of TNM stages III and IV and the alpha group, TNM stages I and II. As shown in Table 5, survival rates at 5 years within the newly created composite stages were as follows: alpha, 90/112 (80%); beta, 33/54 (61%); and gamma, 4/27 (15%). The results were statistically significant, and adjacent stages showed a chi-square for linear trend value of 42.2 with P less than 0001.

Quantitative Comparison of Staging Systems

The purpose of staging is to divide a large, usually heterogeneous group into smaller subgroups that are externally disparate but internally homogeneous with respect to outcome. The qualitative comparison of different staging systems is best done with "face validity" or "common sense" [40], whereas statistical scores and tests are used for the quantitative evaluation of each system's mathematical accomplishments [41].

Scores and Tests Used for Dichtomous Outcomes

When evaluating different staging systems or the same staging system in different populations, one must keep in mind that the level of "performance" of the system depends largely on the variance of the predictors and the rate of development of the outcome event in each study population. (Several excellent sources are available for readers interested in learning more about the quantitative evaluation of staging systems)[42-46]. Some quantitative scores and tests used for dichotomous outcomes (ie, survival) are described below.

Monotonicity of Survival Gradient--The survival gradient is monotonic if each of the successive subgroups has a consistently lower (or higher) survival rate than the preceding group.

The range of survival gradient is the difference between the highest and lowest survival rates in the staging system. A wide range is obviously desirable.

Proportionate Reduction in Predictive Errors--A staging system divides a population into unique prognostic strata. When the outcome of interest is a dichotomous event, all the members of a stratum are "predicted" to have attained or not attained the target event, depending on whether the stratum rate is above or below 50%. For instance, if the survival rate for a particular stage were higher than 50%, each member of this stratum would be predicted to survive.

If e(i) is the number of errors in each stratum, the concordant error rate is Sigma e(i)/N, where N equals the number of individuals in the total, unstratified, population. The smaller this rate, the more concordant is the fit of the staging partition. The total improvement in congruent fit can be expressed as the proportionate reduction in predictive errors. Thus, if E is the number of congruent errors in the total, unstratified data, the total improvement in congruence is (E - Sigma e(i)/E. The larger the proportionate reduction in predictive error, the better is the staging system.

Proportionate Reduction in Variance-The proportionate reduction in variance refers to the proportion of group variance in the original population that was later reduced by the division into stages. The score ranges from 0 to 1, and higher values represent better achievements for staging systems having the same number of categories.

The formula for calculating the proportionate reduction in variance is expressed as (NPQ - Sigma n(i)p(i)q(i)/NPQ, where N, P, and Q are the total number of patients, overall survival rate, and overall death rate, respectively; and n(i), p(i), and q(i) are the number of patients, survival rate, and death rate, respectively, for patients in each stage of the system (i).

Chi-Square for Linear Trend-This score represents a test of the linear monotonicity of the survival rates within ordered categories. It is not equivalent to the standard chi square. For comparative purposes, the higher the value of chi squared (lt), the better. For nonparametric data, the Mann-Whitney U-test should be used.

The c-statistic is derived from measures of sensitivity and specificity and is equal to the area under the receiver operating characteristic (ROC) curve. The larger the number, the better is the staging system.

Results of Quantitative Comparison

The results of the quantitative evaluation of the TNM and composite staging systems are shown in Table 6. Both systems produced a monotonic survival gradient and distributed the entire population into the various stages fairly equally. For each of the quantitative tests, the composite staging system performed better than the TNM system.

Prospective collection of data from a different cohort of patients is crucial to demonstrate the validity of the proposed composite staging system. In addition, the unique prognostic impact of comorbidity demonstrated in laryngeal cancer must be separately assessed in other anatomic subsites within the head and neck region before a comprehensive comorbidity-anatomic staging system for head and neck cancer can be proposed.

Conclusion

The widespread use of the TNM staging system has helped standardize the classification of human cancers. Nevertheless, clinical factors, such as comorbidity, which are important for classification, prognostication, and evaluation of treatment effectiveness, remain excluded. As stated by Burke and Henson [26],"...to increase our prognostic accuracy, the current system must be enhanced by the creation of a system that contains the TNM variables as well as the new predictive variables."

Taxonomies exist to classify comorbidity, and statistical techniques are available to include multiple variables into a single staging system. The continued exclusion of important clinical factors perpetuates an imprecise cancer classification system, and the inclusion of these variables should be a top priority of clinical researchers.

References:

1. Denoix PF, Schwartz D: Regeles generales de classification des cancers et de presentation des resultats therapeutics. Acad Chirurg (Paris) 85:415-424, 1959.

2. International Union Against Cancer: TNM Classification of Malignant Tumours, Ed 4, pp 1-3. Berlin, Germany, Springer-Verlag, 1987.

3. American Joint Committee on Cancer: Manual for Staging of Cancer, 3rd Ed. Philadelphia, JB Lippincott, 1988.

4. Copeland MM: Clinical staging of cancer for end result reporting, in Clark RL, Cumley RW (eds): Yearbook of Cancer, pp 498-503. Chicago, Yearbook Publishers, 1960.

5. Feinstein AR: Symptoms as an index of biological behaviour and prognosis in human cancer. Nature 209:241-245, 1966.

6. Charlson ME, Feinstein AR: The auxometric dimension. JAMA 228:180-185, 1974.

7. Clemens JD, Feinstein AR, Holabird N, et al: A new clinical-anatomic staging system for evaluating prognosis and treatment of prostatic cancer. J Chronic Dis 39:913-928, 1986.

8. Neel HB, Taylor WF: New staging system for nasopharyngeal carcinoma. Arch Otolaryngol Head Neck Surg 115:1293-1303, 1989.

9. Taylor WF, Ivins JC, Unni KK, et al: Prognostic variables in osteosarcoma: A multi-institutional study. J Natl Cancer Inst 81:21-30, 1989.

10. Peipert JF, Wells CK, Schwartz PE, et al: The impact of symptoms and comorbidity on prognosis in stage IB cervical cancer. Am J Obstet Gynecol 169:598-604, 1993.

11. Karnofsky DA, Burchenal JH: The clinical evaluation of chemotherapeutic agents in cancer, in Macleod CM (ed): Evaluation of Chemotherapeutic Agents. New York, Columbia University Press, pp 191-205, 1949.

12. Stell PM: Prognosis in laryngeal carcinoma: Host factors. Clin Otolaryngol 13:399-409, 1989.

13. Bennett CL, Greenfield S, Aronow H, et al: Patterns of care related to age of men with prostrate cancer. Cancer 67:2633-2641, 1991.

14. Boyd NF, Clemens JD, Feinstein AR: Pretherapeutic morbidity in the prognostic staging of acute leukemia. Arch Intern Med 139:324-328, 1979.

15. Charlson ME, Pompei P, Ales HL, et al: A new method of classifying prognostic comorbidity in longitudinal studies: Development and validation. J Chronic Dis 40:373-383, 1987.

16. Piccirillo JF, Wells CK, Sasaki CT, et al: New clinical severity staging system for cancer of the larynx: Five-year survival rates. Ann Otol Rhinol Laryngol 103:83-92, 1994.

17. Wells CK, Stoller JK, Feinstein AR, et al: Comorbid and clinical determinants of prognosis in endometrial ca. Arch Intern Med 144:2004, 1984.

18. MacDonald I: The individual basis of biologic variability in cancer (editorial). Surg Gynecol Obstet 106:227-229, 1958.

19. Feinstein AR: On classifying cancers while treating patients (editorial). Arch Intern Med 145:1789-1791, 1985.

20. Barr LC, Baum M: Time to abandon TNM staging of breast cancer? Lancet 339:915-917, 1992.

21. Feinstein AR, Wells CK: A clinical-severity staging system for patients with lung cancer. Medicine 69:1-33, 1990.

22. Zelen M: Keynote address on biostatistics and data retrieval. CancerChemother Rep 4:31-42, 1973.

23. Adami HO, Malker B, Holmberg L, et al: The relationship between survival and age at diagnosis in breast cancer. N Engl J Med 315:559-563, 1986.

24. Sigurdsson H, Baldetrop B, Borg A, et al: Indicators of prognosis in node-negative breast cancer. N Engl J Med 322:1045-1053, 1990.

25. Bailey BJ: Beyond the 'new' TNM classification (editorial). Arch Otolaryngol Head Neck Surg 117:369-370, 1991.

26. Burke HB, Henson DE: Criteria for prognostic factors and for an enhanced prognostic system. Cancer 72:3131-3135, 1993.

27. Feinstein AR, Schimpff CR, Hull EW: A reappraisal of staging and therapy for patients with cancer of the rectum: II. Arch Intern Med 135:1454-1462, 1975.

28. Feinstein AR, Schimpff CR, Hull EW: A reappraisal of staging and therapy for patients with cancer of the rectum: I. Arch Intern Med 135:1441-1453, 1975.

29. Feinstein AR, Schimpff CR, Andrews JF Jr, et al: Cancer of the larynx. J Chronic Dis 30:277-305, 1977.

30. Greenfield S, Aronow HU, Elashoff RM, et al: Flaws in mortality data: The hazard of ignoring comorbid disease. JAMA 260:2253-2255, 1988.

31. Kaplan MH, Feinstein AR: The importance of classifying initial comorbidity in evaluating the outcome of diabetes mellitus. J Chronic Dis 27:387-404, 1974.

32. Greenfield S, Apolone G: The importance of coexistent disease in the occurrence of postoperative complications and one-year recovery in patients undergoing total hip replacement: Comorbidity and outcomes after hip replacement. Med Care 31:141-154, 1993.

33. Satariano WA: Comorbidity and functional status in older women with breast cancer implications for screening, treatment, and prognosis. J Gerontol 47:24-31, 1992.

34. Concato J, Horwitz RI, Feinstein AR, et al: Problems of comorbidity in mortality after prostatectomy. JAMA 267:1077-1082, 1992.

35. Greenfield S, Bianco DM, Elashoff RM, et al: Patterns of care related to age of breast cancer patients. JAMA 257:2766-2770, 1987.

36. Satariano WA, Ragland DR: The effect of comorbidity on 3-year survival of women with primary breast cancer. Ann Intern Med 120:104-110, 1994.

37. Breiman L, Friedman JH, Olshen RA, et al: Classification and Regression Trees. Belmont, California, Wadsworth International Group, 1984.

38. Gordon TJ: Hazards in the use of the logistic function with special reference to data from prospective cardiovascular studies. J Chronic Dis 27:97-102, 1974.

39. Concato J, Schwartzman D, Feinstein AR: A comparison of logistic regression and conjunctive consolidation as methods of multivariable analysis (abstract). Clin Res 42:291a, 1994.

40. Feinstein AR: Clinimetrics. New Haven, Yale University Press, 1987.

41. Iezzoni LI: Risk and outcome, in Iezzoni LI (ed): Risk Adjustment for Measuring Health Care Outcomes, pp 1-28. Ann Arbor, Health Administration Press, 1994.

42. Feinstein AR: Clinical biostatistics: XIV. The purposes of prognostic stratification. Clin Pharmacol Ther 13:285-297, 1972.

43. Feinstein AR: Clinical biostatistics: XV. The process of prognostic stratification. Part 1. Clin Pharmacol Ther 13:442-457, 1972.

45. Feinstein AR: Clinical biostatistics: XVI. The process of prognostic stratification. Part 2. Clin Pharmacol Ther 13:609-624, 1972.

46. Kramer MS: Clinical Epidemiology and Biostatistics. New York, Springer-Verlag, 1988.

47. Harrell FE, Lee KL, Califf RM, et al: Regression modeling strategies for improved prognostic prediction. Stat Med 3:143-152, 1984.