The natural history of pancreatic cancer in Manitoba : A population based study

Adenocarcinoma of the pancreas is a virulent malignancy. A retrospective study was undertaken to determine the natural history of the disease in Manitoba, and to examine which facets of the disease might predict survival. A retrospective review of the Manitoba Tumour Registry from January 1987 to December 1989 was performed. Two hundred and fourteen exocrine cases were identified, and known islet cell tumours, lymphomas and sarcomas were excluded. The variables analyzed were age, sex, surgical or radiological stage of disease, pathological differentiation, means of diagnosis and treatment. There were 112 females and 102 males with a median age of 73 years (80% of patients were older than 60 years). The mean survival from time of diagnosis was only six weeks. Univariate analysis identified male sex, advanced stages and poor differentiation as poor prognostic indicators. Multivariate analysis revealed that advanced stage and lack of differentiation were independent prognostic variables. Treatment, which was very selective, was also noted on univariate analysis to be a prognostic indicator but was not used in the multivariate regression analysis. The poor prognosis of all individuals confirms the rationale of trying to develop better treatments for all stages of disease, and stage and pathological differentiation should be taken into account when designing clinical trials.

Manitoba Tumor Registry de janvier 1987 a decembre 1989 a done etc entreprise.A PENOCARCINOMA OF THE PAN- creas is a virulent malignancy with an exLremcly shnn ~urvival, even in what appears en he locally invasive surgically resecrablc cases.From the l 900s Lo the l 970s, there has been a striking rise in the inc idence anJ mortality from r a ncrcatic cancer in bt)th sexes ( l ).This increased incidence occurred in mosL European counLri es and in the United Stales (I), but seems to have plateaued m ma ny countries for unknown reasons (2,3), and has remained sleady throughouL the l 980s.Data from Cn ncer Statistics (Canada) fo r the fim half of tht.: I 980s also show a remarkably constant number of abouL 2400 cases per year in this country (3 ).Disease distrihuLion in Canada reveals a relative ho mogeneity of cases across the country Wllh no obvious h igh or low inc idence areas ( 3).Ir was th us considered that any figures derived for Man iroba should be representative fo r all of Canad a.

Depanmem of
There have heen several epidemiological studi es of pancreatic cancer, ofte n comhinecJ with surgical approaches to identify important prognostic factors ( 2,(4)(5)(6).However, the present a uthors were about to perform a phase 11 drug study (Lederle ) using survival as an endpo int.This study aimed to delineate carefully the natural history of pancreatic cancer in the population, in order w ascertain which factors may he important for prognosb and sur-nesid ioblastomcs, les lymphomcs et les sarcomes furent exclus.Les variables analysees ont e te: l'agc, le sexe, le stade chirurgical ou radiologique de la maladie, la d ifferen t iation pathologique, les techniques diagn ostiques et le traitement.L'etude comporta it L 12 femmes et 102 ho mmes, dont l'age moyen etait de 73 ans (80 % des patients avaient plus de 60 ans).La survie moyenne apres le diagostic etait d'env iron six mo is.Une analyse univariee a don ne comme indicateurs du pronostic: les sujets de sexe masculin, le stade avance de la maladie et une d iffe rentiation peu ma rquee.L'analyse mult ivariee quan ta elle a revele que le stade avance de la maladie et !'absence de differe ntiation etaient des variables independan tes du pron ostic.Le traite ment, qui fut tres selectif, fu t egalement considere parmi L es indicatcurs du pronostic clans !'analyse univariee mais ne fut pas utilise clans !'analyse mult ivariee regressive.Le pronostic sombre pour tous Jes sujets justifie done la rech erche de meille urs traitements pour toutes les etapes de L a maL adie.De meme, la d ifferentiation.au plan de la path ologie et d u stade de la maladie doit auss i en trer en ligne de compte lors de la modeLisation des essais clin iques.

PATIENTS AND METHODS
The Manitoba Tumour Registry provided a three-year (J anuary 1987 to December l 989) roster of 2 14 cases o( all exocrine pancreatic carcinoma, (biopsy, cyrology o r radiological).Known islet cell tumours, lymphoma, and sarcomas were excluded from the analysis.No atte mpt was made to restrict entry, so that even indivicluab who were diagnosed ac death were included in this analys is.T he dace of the d iagnostic procedure ( usually pa thological but in some circ umsta nces only radi ological), was taken as the date of diagnosis.Histo logical or cyto logical diagn osis took precedence over radiological diagnosis, so that imaging techniques were onl y used fo r di agnosis if t issue or cytology was unava ilable.
The date of d iagnosis a nd last follow-up or death was recorded.The cases were analyzed fo r the fo llowing vari ables: age, sex, surgical or radiological stage of d isease, pa thological diffe rentiation , means of diagnosis anJ treatme nt.Surgical a nd radio logical stages were lumped toge the r fo r the purposes of analys is as survival was not differen t accord ing to eithe r method of staging.
Surviva l plots we re determined by the actuarial I ifc-ta ble method and were compared in a univariate an alysis P • 0.      Multivariate regression analysis was performed on parameters which were significant on univar iate ana lysis.

RESULTS
Two hundred and fourteen case::.were provided, l l 2 females and I 02 males, a t va riance with the usual sex Jisrribution.The median age was 73 years (range 40 co 96).Eighty-eight per cent were ove r 60 years of age at diagnosis.The mean surviva l fro m time of diagnosis was o nl y six weeks (Figure l ).The sho rtest surviva l was less than o ne week and the longe t survivor lasted 78 weeks.
Univariate ana lysis was carried o ut using the variables mentio ned.Age was not found to be a significant factor in survival.Female survival was slightly but significantly (P=0.05)greater than ma le s urvival (Figure 2).The median surv ival offemales was 6.5 weeks compared with a medi an male survival of on ly three weeks.O ne hyporhes1s was that ind1 v1dua b who were diagnosed solely hy rad iological meam wou ld not actua lly have pancreatic cancer.T herefore, the authors wanted to measure the survival of the two groups (radio logical diagnosis 6 l cases and pathological diagnosis I 5 3 case~).believing that those with a rad1ological diagnosis wou ld su rvive longer (as ,ome of the individuals migh t have hcn1gn disease, pancreatitis, etc). 1 lowever, there was no significant difference in t he survival of patients with ,1 tissue d iagnosb from those diagnosed by im aging procedu res (Figu re 3).
The authors a lso wbhed w exam ine whether patient~ with more poorly tliffe remiated tun1l)urs had poorer survival than those who appeared to have lower grade characteristics, as viewed hy the interpreting pathologisc.No a ttempt was made to rev iew a ll the pat hological tissue.Nm all subm itted specimens had been graded as to cel lular differentiation.In those t hat were (88 cases) (Figure 4), pat ients with some tlegree of different1nt ion (39 cases were we ll, or moderately differentiated) ou tl ivetl those wi th poorly d ifferentiated o r anaplastic rumours (47 case) (P=0.001).The other cases either had not been biopsied (61 cases) or no gradation of the cance r was reported (67 cases).
Stagin g was assigned accordi ng to the American Joint Committee fo r Cancer Staging a nd End Rebults Reporting (1 ), (T able l) In patients fo r whom assigning the stage from the information was poss ible (only two of The num ber of patie nts receiv ing any defi n iti ve treatment fo r cure was ex tremel y sma ll {8%).O nly six of 214 had definitive surgery, which was labelled non palliat ive by the surgeon .T he reason fo r this dist ributio n is unknown.U n iva riate ana lys is shows that an y ex tra treatment, whether radi otherapy o r chemo therapy, added to the in itial surgery, increased survi va l (Figure 6 ).However, the numbe rs a rc small (only 21 of 214 patients) and muc h presclec tion is st rongly suspec ted .Multiva riate regressio n analysis: Cox regression ana lys is revealed tha t onl y stage and histology remained independen t prognostic variables, wit h sex d is-appearing.Treatmen t was no t used in the model because of th e suspected select io n .

DISCUSSION
A po pu lation -based study of Mani toba pa tients with pa ncreat ic carci noma was ana lyzed to dete rmine survival and prognostic vari ables.Stage a nd histo logy were fo und to be indepe nde nt prognostic va riables a nd sho uld he take n into consideratio n in any th erapeuti c trial in vo lving this disease.This is true even though the patho logy was not revieweJ .
Of great con cern is th e extre mely short surviva l ch at was seen in this d isease in Ma nitoba with th e relat ively fe w definitive treatment interventions.Most other seri es report me<lian survivals of six co 22 mo nths (1,4-6), while the med ian survival for the present group was o n ly six weeks.Even patients with the best prognos is haJ a median survival of on ly 10 to J 2 wee ks, although the tails of the surviva l curves flatte n o ut after the init ial rapid drop off rate.It is poss ible this totall y unselected series baseJ o n a tumour regist ry presen ts a true picture of a viscious disease.By the time the patient h as a consultat io n with a surgeo n , gastroenterologist o r oncologist, they have passed o ne maj or surv iva l hurd le and t hus will survi ve lo nger.If we exclude the first six to e ight weeks of rapid death to allo w 'consultat ions' the c urves resemble othe r series.Using that maneuvre , median survival of the present patients is nine to IO months, approxima ting wha t is fo und in the lite rature.Thus, sho rt survival is likely J ue to the fac t tha t a sizeable proport io n of patients were di agnosed close co death.C learly, survi val coulJ be 'extended' fro m symptoms, rather than patho logical o r radi o logical diagnosis.These are probably th e reasons th e surviva l is so sho rt.T h is stud y unde rl ines th e impo rtance of careful study of prognost ic variables in any case con trol study, and the long known Jiffi culry of histori cal conrrols.

WEEKSFigu re 1 )
Figu re 1) Pancreatic cancer survival of a series of 2 I 4 parientsfrom the Manito ba Tumour Registry.a Alive; Ai,•ows confide nce limits

TABLE 1
Staging of pancreatic cancer*