Colorectal cancer—a multivariate analysis of prognostic factors

European Journal of Surgical Oncology 1996; 22:592-597

Colorectal cancer

a multivariate analysis of prognostic factors

U. Wolters*, H. Stiitzert, H. W. Keller*, U. SchrOder* and H. Pichlmaier* *Department of Surgery, University of Cologne, and "~Department of Medical Documentation and Statistics, University of Cologne, Germany

The data from 1050 patients who had undergone coloreetal carcinoma resection at the University of Cologne between 1976 and 1990 were studied. The aim of the study was to determine the concomitant effects on survival of several patient characteristics (sex, age, turnout localization, blood transfusion) and histopathological variables (Dukes' staging). We first e~lculated survival rates, both including and excluding post-operative mortality. We set up a hierarchical log-linear rdodel for the detection of relationships between selected crossclassified categorial variables. We then used Cox's proportional hazard regression method to study the relationship between survival and different prognostic patterns. Dukes' staging was shown to be a highly discriminating factor in survival (P<0.001). Survival rates were better in women (P<0.001), and better for younger patients (<70 years; P<0.001). Tumour site (colon; P = 0.0362) and blood transfusion (P=0.0857) also correlated with survival. Key words: colon carcinoma; risk factors; long-term survival; cancerJn the elderly; sex-specific survival.

Introduction The prognosis of patients with colorectal cancer mainly depends upon tumour stage. Other factors such as advanced age, sex, tumour site, ~-3 blood transfusions, 4 and septic complications of surgery can influence survival. Univariate analyses can be misleading. Multivariate analyses, which can demonstrate the simultaneous effects of many prognostic factors. This report analyses the survival data of 1050 patients in order to identify and validate the prognostic factors with colorectal cancer related to survival.

Patients and methods We studied data from all patients (n = 1050) operated on in our department for colorectal cancer between January 1976 and December 1990. Standard follow-up was carried out between January and July 1991. Variables such as advanced age (over 70 years), sex, tumour stage (modified Dukes' classification), tumour localization (grouped for analysis of prognosis: right or left colon/sigma or rectum), peri-operative blood transfusion (no/yes), and post-operative complications (none/

Correspondence to: U. Wolters, Chirurgische Universit/itsklinik K61n, J. Stelzmannstr. 9, 50924 K61n, Germany. 0748-79831961060592+ 06 $12.00/0

moderate*/severet) were studied. We used standard univariate and multivariate analyses to analyse the data. We calculated observed survival rates 5 including postoperative mortality. To evaluate long-term prognosis, excluding peri-operative events, we also calculated survival rates for those patients alive at 30 days following surgery. For univariate survival analyses the stated P-values are based on Breslow's version of a generalized non-parametric rank test. 6 Expected survival rates were used to adjust for ageand sex-related risk of death from the general population proposed by Cutler et al.7 For each year of follow-up, and assuming equal person-years at risk, the expected death rate of observed patients was computed by cumulative survival rates from general population life-tables (published in Annual Reports of the Statistisches Bundesamt for die Bundesrepublik) matching age and sex for each of our pati~ts. The effect of increasing survival expectancy with time has been taken into account using different life-tables~ for follow-up years occurring during the two calendar periods defined by the cut-off point 1980. Computations were done by a computer program 9 developed by one of the authors. Because of possible confounding effects of peri-operative prognostic variables we also set up hierarchical log-linear" models.~-4"L°

* Moderate: pleura effusion/wound infection/suture insufficiency/ catheter sepsis/abscess (but not intra-abdominal)/haematoma/ aseites/thrombosis/retention of urine. t Severe: pneumonia/p.o, bleeding/sepsis/peritonitis/striking cardiac or pulmonal complication/hepatic or renal insufficiency/ ileus/embolisndreanimation/intra-abdominal abscess. © 1996 W.B.SaundersCompany Limited

Colorectal cancer---multivariate analysis

593

1.0

0.8

!~ 0.6

0.4

0.2

_

I

I

12

24

J

36

48 60 72 84 Time after surgery (months)

96

108

120

Fig. 1. Observed survival stratified by sex and age. (-O-=male patients ~<70 ~ears (n=329); ( - 1 - = m a l e patients >70 years (n= 155); ( - O - = female p~tients ~<70years (n=329); (-I--l-=female patients >70 years (n= 172).).

1.0

0.8

i~ o.6

0.4 t~ 0.2

0

I

1

I

12

24

36

I

I

I

I

48 60 72 84 Time at~er surgery (months)

I

I

I

96

108

120

Fig. 2. Expected survival stratified by sex and age based upon standard life tables. -O---patients aged ~<70years (n=658) (-O-=male (n=329); - - - O . . . . female (n=329)). -I'-I-= patients aged >70 years (n=327); ( - 1 - = m a l e (n= 155); - - - I I . . . . female (n= 172)).

We used Cox's proportional hazards regression method" for age, sex, stage and localization of tumour, blood transfusion and peri-operative complicationS. We also modelled interactions of a second order between some of these factors. Tests based on likelihood ratio methods and large-sample normal approximation were used to examine the significance of some subsets of regression coefficients and to compare the model fits with submodels restricted to smaller sets of prognostic variables. P-values are given without adjustment for multiple significance tests.

Computations were made using the BMDP® statistical package.t-"13

Results

A total of 1050 patients (mean age 64.6 yrs, range 23-94) underwent operations for colorectal adenocarcinoma between 1 January 1976 and 31 December 1990 (Table 1). Sixty-five patients (6.2%) died within 30 days after operation;

U. Wolters et al.

594 1.0 ¸

0.8 laa

!

"~ 0.6

~

0.4

0.2

0

12

24

36

iiiii .. iii

48 60 72 84 Time after surgery (months)

96

108

120

Fig. 3. Modelled (-II-=Dukes' A; - O - = D u k e s ' B; -A---Dukes' C; - ~ - = D u k e s ' D) and observed (-[-I-=Dukes' A; - O - = D u k e s ' B ; - A - = D u k e s ' C;-<>-=Dukes' D) survival (based on Cox's proportional hazards model)•

Table 1. Patients' general character~l"cs

Sex

Age Dukes'

All cases Peri-operative mortality Male Female <~70 >70 A B

C D Missing Localization

Blood transfusions Complications

Right colon Left colon Sigmoid colon Rectum No Yes None Moderate Severe

these patients' data were excluded from the analysis of cancer-related risk factors in long-term survival.

Univariate analysis Univariate analysis of patients' age, sex, tumour stage, blood transfusions and the occurrence of post-operative complications revealed statistically significant associations with long-term survival (Table 2). An association could not be detected with tumour location. No significant difference between rectum amputation and resection was seen. At this stage, stratified analyses suggested that long-term prognosis was associated with patient's sex and age (Fig. 1). This may be due to sex-related life expectancy as estimated from general life-tables (Fig. 2).

n = 1050 65 520 530 691 359 226 306 323 194

n=985 Patients alive 30 days aNer surgery 484 501 658 327 213 288 308 176

1

173 108 300 469 484 566 624 260 166

164 100 279 442 474 511 616 256 113

Multivariate analysis A Cox regression model was set up to describe and quantify the influence of these factors on long-term survival. Only patients surviving more than 30 days were included in this proportional hazards analysis. We used a backward elimination technique ~t3 starting from a set of eight risk variables (age ~<70 vs >70), sex, tumour stage (representing the four Dukes' stages (A to D) by three binary dummy* variables D1,D2,D3), localization • of tumour (right or left colon vs sigmoid colon or rectum), peri-operative blood

*That is: Dukes' stage A is represented by D I = D 2 = D 3 = 0 , whereas Dukes' stage C is represented by DI = D 2 = 1 and D3=0, Dukes' stage D is represented by D 1= D2 = D3 = 1.

Colorectal cancer--multivariate analysis

595

Table 2. Univariate survival analysis

Risk variable

All patients n= 1050 (I case missing for Dukes) Survival Category

Sex Age Dukes

Localization Blood Complication

Male Female ~<70 >70 A B C D R. col./I, col. Sigma/rectum No Yes None Moderately/severe

N. of pats 520 530 691 359 226 306 323 194 291 769 484 566 884 166

Median [months] 38.8 -t- 5.0 68.7 +__10.6 63.7+7.2 40.2 -I-4.8 > 120 101.2+24.6 39.8-t-5.0 8.3+0.8 58.6+17.1 51.3 -I-4.3 71.2+15.1 40.6 + 5.2 63.2 -I-5. I 11.7 _ 3.5

5-yr rate [% +SE] 41.4 ___2.5 52.2 -I-2.4 51.1 +2.1 38.6 + 3.0 72.7-1- 3.3 61.0+3.2 40.2+3.2 2.1 _ 1.3 49.7-t-3.3 45.9 ___2. I 52.8-1-2.6 42.2 -I-2.3 51.2 + 1.9 25.9 __+3.5

P [Breslow] <0.000 I 0.0014 <0.0001 0.7733 <0.0001 <0.0001

Table 3. Relative risks Prognostic variable (including interactions) Dukes (*--,* Age, site, blood) Sex Age (*--,* Dukes) Tumour site (,--** Dukes, complication) Complication (~* Tumour site) Blood transfusion (* Dukes)

Most favourable* Most unfavourable* Stage A Stage D Female Male ~<70 >70 Proximal: right or left colon Distal: sigmoid or rectum Non or moderate Severe No Yes

Relative riskt

Risk ratio~

0.39 6.54 0.84 1.20 0.84 1.42 0.86 1.06 0.99 1.03 0.94 1.05

16.8 1.4 1.7 1.2 1.04 I. I

* Variables representing the interaction term(s) were set to the mean value observed for stratum named in the respective line of this column; therefore values stated for relative risk or risk ratio may differ from those computed on base of the named risk factor alone; the definition of interactions cited is given in Table 4. t Related to mean covariate pattern (modelled survival for complete collective). :~Of most unfavourable to most favourable as named in respective lines.

transfusions, complications. We explored relationships between patterns of risk factors and long-term survival described in terms of the regression coefficients of the appropriate proportional hazards model, and relative risks of selected patterns of prognostic variables (Table 3). Additional results of this modelling process, including interaction terms, are given in Table 3. Calculations were based on the formal assumption that the other risk factors are at their average value. Contrasted with the results of univariate analyses, long-term survival was alsd influenced by tumour site (colon vs sigrnoid or rectum) (P=0.0362). A slight influence (P=0.0857) of peri-operative blood transfusion on long-term survival could be detected by this modelling process, the relative risk attached to perioperative blood transfusions being only 1.1. The most striking influence on long-term survival is given by the Dukes' level of staging for the tumour (Fig. 3). The

relative risk related to tumours classified as Dukes' D being about 17-fold that for Dukes' A (Table 3). The influence of blood transfusions is least important. Female patients generally had a better prognosis than males, especially in younger patients, the relative risk of males >70 years being about 2.7-fold that for females />70 years.

Discussion

Our results suggest that survival rates in colorectal cancer depend on tumour stage, on sex, age, tumour location and peri-operative blood transfusions. We confirm that tumour stage is the most important prognostic factor. Reports about the prognostic significance of tumour site

596

U. Waiters et al.

in colorectal cancer are controversial. Right colon cancer is found to be favourable by Halvorsen 14-~6and Wolmark,-" sigmoid colon cancer was shown to be a favourable diagnostic indicator by Shepard n and Stewart. t8 Other studies do not confirm any tumour site as being favourable. u'14-16'lg'-'° The multifactorial analysis of Kune 2t and Enblad 22confirmed better life expectancy in right colon cancer. In our study, univariate analysis did not disclose any prognostic relevance of tumour site, but this was demonstrated by multivariate analysis. These differences might explain the controversial reports of other investigators• We find that the importance of tumour site for long-term survival in colorectal cancer depends on tumour stage. In early cases (Dukes' A) long-term survival for sigrnoid colon and rectal cancer did not markedly differ from right and left colon cancer, unlike the more advanced tumour stages (Dukes' B, C). In o ~ analysis, patient's age, especially in the advanced cancer stages, directly affected survival. The elderly (>70 years) had a poorer outcome than younger (~<70 years) patients? 9-2j'-'4.-'s T-lymphocyte function and interleukin-2 production is depressed in older people. '-3 Multi-variate analyses 26 have demonstrated a significant correlation of patient age with long-term survival. Surprisingly, patients who are still alive 10 years after surgery had a better prognosis than expected when compared with the matched individuals from general lifetables. Female patients generally had a better prognosis, independent of age. Other reports describe contradictory res u i t s. -'o-~.2s.2s.-'9 There has been discussion about the consequences of peri-operative blood transfusion. Some reports describe a shorter 4'3°-33and some a longer life3"~39expectancy following blood transfusion. For kidney transplantation, better graft survival was demonstrated in transfused patients by Opetz ~° in 1980. It seems that the effect depends on the quantity, leading to increased activity of suppressor T-lymphocytes 4t and decreased activity of natural killer cells. 4"- In our investigations transfused patients displayed shorter life expectancy than non-transfused patients. All patients who died after the operation had blood transfusions. Some authors suggest that peri-operative blood transfusion is nothing but an expression of other problems such as mechanical pulmonary ventilation and the course of the operation itself. 43'44 We conclude that the presentation of prognostic relevant factors of colorectal cancer requires a multifactorial observation, which most reports do not sufficiently respect. This might be the best explanation for the many contradictory results presented elsewhere in the literature.

References

1. Kashtan H, Werbin N, Aladjem D, Barak Y, Wiznitzer T. Right and left colon carcinoma: a retrospective comparative study. J Surg Oneol 1987; 35: 245-8. 2. Wolmark N, Wienand, S, Rockette HE, Fisher B, Glass A, Lawrence W, Lerner H, Cruz AB, Volk H, Shibata H, Evans J, Prager D. The prognostic significance of tumor location and bowel obstruction in Dukes B and C colorectal cancer. Ann Surg 1983; 198: 743-52.

3. Miller W, Ota D, Giacco G, Guinee V, Irimuras T, Nicolson G, Cleary K. Absence of a relationship of size of primary colon carcinoma with metastatic and survival. Clin Expl Metastasis 1985; 3: 189-96. 4. Foster RS, Costanza MC, Foster JC, Wanner MC, Foster CB. Adverse relationship between blood transfusion and survival after colectomy for colon cancer. Cancer 1985; 55:1195-201. 5. Kaplan EL. Meier P. Nonparametric estimation from incomplete observations, d Am Star Ass 1958; 53: 457-81. 6. Breslow N. A generalized KruskaI-Wallis test for comparing k samples subject to unequal patterns of censorship. Biometrika 1970; 57: 579-94. 7. Cutler SJ, Griswold MH, Eisenberg H. An interpretation of survival rates in cancer of the breast. J Nat Cancer hlst 1957; 19: 1107-17. 8. Statistisches Bundesamt ft~r die Bundesrepublik Deutschland. Statistisches Jahrbuch. Ausgaben 1971, 1985. 9. Sttitzer H. EXSUAN. A program for computation of expected survival using general population based life tables (personally committed). 10. Christensen R. Log-L#~ear Models. New York: Springer, 1990. I1. Cox DR. Regression models and life tables. J Roy Stat Sac 1972; 34B: 187-220. 12. BMDP Statistical Software, Inc. Los Angeles, CA, 1990. 13. Dixon WJ, Brown MB, Engelman L. BMDP Statistical Software Manual. Vol 1,2. University of California Press, Berkeley, 1990. 14. Halvorsen TB, Seim E. Tumour site: A prognostic factor in colo~ectal cancer? A multivariate analysis. ScandJ Gastroenterol 1987;"~." 124-8. 15. Wood DA'-,,.Robbins GF, Zippin C, Lum D, Stearns M. Staging of cancer of the colon and cancer of the rectum. Cancer 1979; 43: 961-8. 16. McDermott FT, Hughes ESR, Pihl E, Milne BJ, Price AB. Comparative results of surgical management of single carcinomas of the colon and rectum: a series of 1939 patients managed by one surgeon. Br J Slo'g 1981, 68: 850-5. 17. Shepard RJ, Jones JSP. Adenocarcinoma of the large bowel. Br J Cancer 1971; 25: 680-90. 18. Stewart RJ, Robson RA, Stewart AW, Stewart JM, McBeth AAG. Cancer of the large bowel in a defined population: Canterbury, New Zealand 1970--4. Br J Surg 1979; 66: 309-14. 19. Hannisdal E, Thorsen G. Regression analysis of prognostic factors in colorectal cancer. J Surg Oncol 1988; 37: 109-12. 20. Wiggers T, Arends JW, Volovics A. Regression analysis of prognostic factors in colorectal cancer after curative resections. Dis Colon Rectum 1988; 31: 33-41. 21. Kune GA, Kune S, Field B, White R, Brough W, Schellenberger R, Watson LF. Survival in patients with large-bowel cancer. Dis Colon Rectum 1990; 33: 938-45. 22. Enblad P, Adami HO, Bergstrom R, Glimelius B, Krusemo UB, Pahlman L. Improved survival of patierits with cancers of the colon and rectum? J Natl Cancer lnst 1988; 80: 586-90. 23..Gillis S, Kozah R, Durante M, Wehsler ME. Immunological studies of aging. Decreased production of and response to T cell growth factor by lymphocytes from aged humans. J Clin h~vest 1981; 67: 937--44. 24. Bossola M, Sofa L, Ratto C, Bellantone R, Doglietto GB, Crucitti A. Colorectal cancer in patients 40 years of age and younger. Ita! J Surg Science 1989; 19: 361-7. 25. lrvin "IT, Prognosis of colorectal cancer in the elderly. Br J Surg 1988; 75: 419. 26. Goh HS, Goh CR, Rauff A, Foong WC. Clinico-pathological prognostic factors of large bowel cancer in Singapore: a multivariate analysis. Ann Acad Med 1987; 16: 437--40. 27. Svendson LB, Sorensen C, Kjersgaard P, Meisner S, Kjersgaard J. The influence of age upon the survival after curative operation for colorectal cancer. Int J Colorect Dis 1989; 4: 123-7. 28. Gril~n MR, Bergstralh EJ, Coffey J, Beart'RW, Melton JL. Predictors of survival after curative resection of carcinoma of the colon and rectum. Cancer 1987; 60: 2318-24. 29. Stower MJ, Hardcastle JD. The results of 1115 patients with colorectal cancer treated over an 8-year period in a single hospital. Fur J Surg Oncol 1985; 11: 119-23.

Colorectal cancer--multivariate analysis 30. Van Lawick Van Pabst WP, Langenhorst BLAM, Mulder PGH, Marquet RL, Jeekel J. Effect of perioperative blood loss and perioperative blood transfusions on colorectal cancer survival. Eur J Clin Oncol 1988; 24: 741-7. 31. Liewald F, Wirsching RP, Zulke C, Demmel N, Mempel W. Influence of blood transfusions on tumor recurrence and survival rate in colorectal carcinoma. Eur J Cancer 1990; 26: 327-35. 32. Wirsching RP, Demmel N, Liewald F, Mempel W, Zwingers N. EinfluB der Bluttransfusion auf Tumorriickfall und Uberlebensrate beim colorectalen Karzinom. Chirurg 1988; 59: 647-53. 33. Beynon J, Davies PW, Billings P J, Channer JL, Protheroe D, Umpby HC, Mortensen N J, Williamson RCN. Perioperative blood transfusion increases the risk of recurrence in colorectal cancer. Dis Colon Rectum 1989; 29: 975-8. 34. Vente JP, Wiggers Th, Weidema WF, Jeekel J, Obertop H. Perioperative blood transfusions in colorectal cancer. Eur J Surg Oncol 1989; 15: 371-4. 35. Klingler K, Zhang X, Menghini T, Metzger U, Largiader F. EinfluB von Bluttransfusionen aufdas Oberleben von Patienten Mit kolorektalen Karzinomen. Heh, Chh" Acta 1988; 55: 685-8. 36. Weiden PL, Bean MA, Schultz P. Perioperative blood transfusion does not increase the risk of colorectal cancer recurrence. Cancer 1987; 60: 870-4. 37. Tartter PI. Perioperative blood transfusion and colorectal cancer recurrence. J Surg Oncol 1988;'39: 197-200.

597

38. Jakobson EB, Eickhoff JH, Andersen J, Lundvall L." Perioperative blood transfusion and recurrence and death after resection for cancer of the colon and rectum. Scand J Gastroenterol 1990; 25: 435--42. 39. Bentzen SM, Balslev I, Pedersen M, Teglbjaerg PS, HanbergSorensen F, Bone J, Jacobsen NO, Sell A, Overgaard J, Bertelsen K, Hage E, Fenger C, Kronborg O, Hansen L, Hostrup H, Norgaard-Pedersen B. Blood transfusion and prognosis in Dukes B and C colorectal cancer. Eur J Cancer 1990; 26: 457-63. 40. Opelz G, Mickey MR, Terasaki PI. Blood transfusions and kidney transplants: remaining controversies. Transplant Proc 1981; 13: 136-41. 41. Marquet RL, Heystek GA. Induction of suppressor cells by donor-specific blood transfusion and heart transplantation in rats. Transplantation 1981; 31: 272-4. 42. Gascon P, Zoumbos NC, Young NS. Immunologic abnormalities in patients receiving multiple blood transfusions. Ann bltern Med 1984; 100: 173-7. 43. Riddle PR. Disturbed immune reactions following surgery. Br J Surg 1967; 54: 882-6. 44. Jubert AV, Lee TE, Hersh EM, McBride CM. Effects ofsurgery anesthesia and intraoperative blood loss on immunocompetence. J Surg Res 1973; 15: 399-403. Accepted for publication 20 July 1996