Clinical and Histopathological Prognostic Factors in Chondrosarcomas

Purpose. In an attempt to identify clinical and histopathological factors of prognostic importance in chondrosarcomas, 115 cases of malignant and borderline chondromatous tumours were reviewed. Patients/methods. Histopathological features tested for prognostic information as well as reproducibility included cellularity, nuclear pleomorphism, multinucleated cells, mitotic activity and grade. Eleven patients had a biopsy only, and a short survival (median 2.0 years); these were excluded from further analysis. The remaining 104 patients who had received intended curative treatment had a median survival of 14.7 years. Results. In univariate analysis, tumour size, extra-compartmental growth, surgical margin and sex were significantly correlated to recurrence-free survival (RFS); sex was marginally significant while age, site and pathological parameters were not significant. Overall survival (OAS) was likewise found to be independent of pathological features as well as site, size and surgical margin; but age, sex and extra-compartmental growth were statistically significant. However, when the same parameters were entered into a stepwise Cox (multivariate) analysis, only surgical margin, cellularity and pleomorphism were significantly related to RFS; margin, grade, pleomorphism and age to OAS. Overall inter-observer agreement on grade was relatively low: 0.54, with a Kappa value of 0.32. It was not better for the other histological parameters, with the exception of the mitotic count. However, acceptable values were achieved when the material was divided into low-grade (grade I and below) vs high-grade (grade II and III) lesions: overall agreement 0.79, Kappa 0.56. Discussion. Although the grading of chondrosarcomas is in need of improvement, its replacement by semiquantitative evaluation of individual histopathological parameters as performed in this study offers no advantage. Among the clinical parameters, only the adequacy of the surgical treatment and the patient's age appear to be important.


Introduction
C hondrosarcom as follow osteosarcom as as the second m ost com m on prim ary m alignancy of bone. U nlike the latter, there has been no real advances in the treatment of chondrosarcom as through the twentieth century. T he disease is remarkably resistant to both radiotherapy and chem otherapy, and surgery is still the m ainstay of the treatment. 1 Since chondrosarcom as can vary in behaviour between extrem ely low grade, slowly grow ing expansive lesions and highly aggressive, invasive and m etastasizing tum ours, the surgeon has to balance between the necessity to achieve an adequate surgical m argin and the wish to avoid an unnecessary and possib ly disab ling operation. W hile clinical and radiological ® ndings are helpful when judging the aggressiveness of the tum our, the histological grade is often the m ajor factor in determ ining the surgical treatment. Custom arily, three-grade system s are used, with grade I indicating a slowly grow ing, expansive lesion and grade III signifying a locally aggressive tumour with a high risk of recurrence and m etastasis. 2± 4 However, the grading of chondrosarcom as is assum ed to be highly subjective and dif® cult even for pathologists with experience in bone pathology. 5 T his is due partly to the rarity of these tum ours, and partly to a weakness in the grading system s themselves. Apart from subtle differences between them, they are also m ainly descriptive, i.e. they describe the typical grade I, II and III lesions but do not de® ne clear boundaries betw een them , nor do they explain how the individual features should be weighted. 6,7 W hile this m ay not be a problem within individual institutionsÐ where grading is perform ed by a single pathologistÐ it does m ake comparison of treatment results betw een the centres unreliable.

Patients
T he m aterial consists of 115 patients treated for histologically con® rm ed chondrosarcom as at the Soft T issue and Bone T um our Centres in Copenhagen and Aarhus in the period 1965± 1994. Excluded w ere extraskeletal, m esenchym al and so-called dedifferentiated chondrosarcom as. Included were a few cases where the original pathology report had given a diagnosis of borderline chondrom atous tum our (`possible chondrosarcom a' , etc.). The following clinical param eters were selected for analysis: age, sex, tumour site and size, and surgical m argin. M argin was classi® ed according to E nneking. 8 The presence or absence of extracom partm ental extension was evaluated on the com bined evidence of clinical ® ndings, radiology and pathology. Survival was calculated from the day of adm ission orÐ in the case of recurring tum ours or patients referred for other reasonsÐ from the date of the ® rst biopsy indicating m alignancy. End-points w ere clinically or radiologically detected local or distant recurrence (recurrence-free survival, RFS), or death (overall survival, O AS).

H istological analysis
F rom each tum our, one or two H &E-stained slides w ere selected, containing the areas that were judged to be best preserved and diagnostic. T hese slides were then circulated between three pathologists (SD , O M J and T S) and conventional grading was perform ed follow ing the guidelines of G itelis et al., 3 with knowledge of the clinical data (age, site, size). In a second round, four individual histological param eters (cellularity, nuclear pleom orphism , m ultinucleated cells and m itotic count) were graded and scored according to the criteria in Table 1. Area m easurem ents were perform ed w ith the help of ocular counting grids so that the counts could be recalculated per m m 2 ; calibration w as done individually, since the optical properties of the m icroscopes differed. T he param eters had been selected after a prelim inary analysis which included calci® cation, m yxoid changes, cytoplasm ic inclusions and growth pattern; these w ere dropped from further analysis because they gave no prognostic inform ation in unior multivariate analysis (data not shown). Invasive growth (de® ned as demonstrable invasion in bone or soft tissue) had shown borderline signi® cance and was included in the present analysis, although only as evaluated by one pathologist. F or the prognostic analysis, a consensus grade w as agreed upon in all cases with discrepancies; in the individual param eters, the mean value w as used.

Statistical analysis
O nly patients treated with curative intent (n 5 104) were included in the survival analysis. First, a univariate analysis w as perform ed by calculating RF S and OAS with log-rank tests for each of the strati® ed variables. N ext, the sam e variables were subjected to a m ultivariate analysis using Cox' s proportional hazards m odel in a stepwise fashio n, with

Inter-observer variation
T he overall agreement on grade was 0.54 w ith a Kappa value of 0.32 (0.25± 0.39, 95% con® dence all the clinical and individual histological param eters included. p-V alues , 0.05 w ere considered signi®cant; although in the stepwise Cox analysis, the p-valu e for entering param eters was set to 0.05, and for retaining them to 0.10, in order to be able to detect possible borderline signi® cance. T he calculations were done on a computer using thè Stata' statistical software (Stata C orporation, C ollege Station, TX , USA). T he inter-observer agreement on the histopathological param eters was evaluated by calculation of overall agreem ent rates as w ell as the Kappa statistic. 9 Kapp a is 1 if agreem ent is total, and 0 if it is equal to the expected chance agreement.

Patients
T he age and sex distribution of the 115 patients is shown in Fig. 1. The oldest patient was a 83-yearold m an, the youngest a 4-year-old boy. T he latter w as the only patient with Ollier' s disease; four patients had m ultiple cartilaginous exostoses. Table 2 shows the location of the tum ours. T um our sizes varied between 1.5 and 30 cm (m ean, 8.9 cm) in largest diam eter, and 54% were judged to be extracom partm ental. Eleven patients were not treated w ith curative intent, generally because of poor general condition or inoperability (although the size of their tum ours did not differ signi® cantly from the   Table 5. Best agreem ent is found for grade III tum ours, poorest for the`borderline' category (the negative Kappa value indicates that this diagnosis by one pathologist is certain to be disputed by the others). T he low values for the`benign' category re¯ect the bias of sam pling, since only suspectedly or obviously m alignant tum ours were included; the K appa value w ill be low if the prevalence of a diagnosis is either very low or very high in the sam ple. 10 As an experiment, agreement was also calculated after regrouping into only two categories, low ( # I) and high (II 1 III) grade; this results in an overall agreem ent of 0.79 w ith a Kappa value of 0.56 (0.45± 0.66), w hich is in the acceptable range. Table 6 show s the inter-observer agreement for the individual histopathological variab les; the Kappa values are sim ilar to those achieved in grading, only m itotic activity is slightly m ore reproducible.

D iscussion
O ur results stress the overwhelm ing prognostic importance of the surgical margin (Fig. 2), a ® nding which is in accordance with the results of other investigators. 2 is likewise a bad om en, highly signi® cant in the univariate test, but surprisingly not so in the m ultivariate analysis± after treatment, the achieved m argin is clearly m ore im portant than local tum our extension. T he site and size of the tum ours are only m arginally signi® cant; the prognostic infor-  (Table 4). O nly by adjusting for surgical m argin and age could the survival curves demonstrate an increasingly poor prognosis with increasing grade (Fig. 3). This is in contrast to the ® ndings of other investigators who in their univariate analyses were able to demonstrate signi® cant correlation between grade and O AS, although RFS app ears to be poorly correlated to grade. 2,3,12,14,15 A reason for our ® nding may be that our grade was a com prom ise betw een three observers; another that we m ay have unduly em phasized nuclear pleom orphism (see later). However, it is worth noting that Burt et al. 11  three app ear to carry prognostic inform ation, although this only becomes evident in the C ox analysis. C ellularity and nuclear pleom orphism seem to be the m ost im portant factors; high cellularity signi® es a high grade of malignancy, as would be expected, 16 but it cam e as a surprise that nuclear pleom orphism was inversely related to the prognosis: the m ore severe the pleom orphism , the better the prognosis. This feature w as constant and could not be explained by covariation with other variables, and it persisted even when the 11 patients with severe pleom orphism were left out from the analysis. It app ears to contrast with the ® nding that nuclear area and DN A content correlate with grade 17,18 and survival. 16 How ever, a high degree of nuclear pleom orphism (as de® ned in this study) m ay not be a reliable indicator of ploidy status, but could also represent a degenerative phenom enon. T hus, nuclear form (as determined by the ratio long axis/ short axis) was show n not to correlate with D N A content. 19 T he cytogenetics of chondrosarcom as are anyway com plex, with no obvious relation between chrom osomal pattern and histologic grade, 20 although the num ber of abnorm alities appears to correlate with grade and prognosis. 21 O ur inter-observer agreement on grade w as rather poor, a result which con® rm s the suspicion that conventionally perform ed grading of chondrosarcom as is highly subjective. T he results are similar to those achieved by other investigators in areas where param eters or entities are less clearly de® ned than generally supposedÐ for instance, som e neurological signs, 22 or subtyping and grading of pulm onary adenocarcinom as. 23 Reducing the num ber of categories im proves agreem ent, as illustrated by the acceptable values achieved in our m aterial by dividing the lesions solely into low-grade and high-grade lesions. This app roach is debatab le, how ever, since it also results in loss of inform ation. 24 In soft tissue sarcom as, inter-observer variation in grading has been reduced to acceptable levels by evaluating histological param eters (differentiation, necrosis and m itotic count) separately. 25 This approach appears to be less useful as regards chondrosarcom as since our results show the inter-observer agreement on the individual histological param eters to be equally poor. O n the other hand, cellularity and nuclear pleom orphism or size are features that can be estim ated morphometrically 18 and thereby m ore objec-tively.
M orphom etry should therefore be investigated as a m eans to reducing inter-observer variation to m anageable levels.
As m entioned earlier, D N A ploidy has been reported to be a signi® cant independent prognostic factor, 15 but determining this requires special equipm ent and expertise in interpreting the histogram s. O ther techniques such as AgN OR counting and im m unohistochem ical staining for p53 or the proliferation m arkers PCN A or Ki67 are tricky to perform on chondrom atous tissue. They require experience and results have generally been correlated to conventional grade, not clinical outcome. 26± 28 However, a recent report by N awa et al. 29 indicates that reactivity w ith the MIB 1 antibody for Ki67 m ay be a useful independent prognostic m arker. Conclusions are dif® cult to draw, though, because of the rather sm all num bers of patients investigated in these studies.
In conclusion, the conventional subjective grading of chondrosarcom as is informative but less than optim al, and im provem ent should be sought for both inter-observer agreem ent and prognostic inform ation. T his will probab ly need the introduction of new param eters or methods in the grading procedure, but the evaluation of such new m ethods or factors purported to have prognostic implication w ill as a m inimum require suf® cient num bers of patients to allow strati® cation for age, treatm ent and grade, best in m ultivariate analyses. To achieve this, it is necessary to centralize diagnosis and treatment of these rare tum ours, and preferably to establish m ulti-centre cooperation.

A cknowledgem ent
T his w ork was supported by a grant from`Hestehandler af Rù nne M . Jensen og hustru Augusta O livia Jensens m indelegat' .