Originl Articles Mitotic Index of Invsive Brest Crcinom Achieving Cliniclly Meningful Precision nd Evluting Tertil Cutoffs John S. Meyer, MD; Eric Costto, PhD; Hns Peter Grf, PhD Context. Mitotic figure counts re relted to brest cncer behvior but hve not been sufficiently reproducible to be ccepted for clinicl decision-mking. Objective. To improve reproducibility nd ccurcy of the mitotic count. Design. Mitotic index (MI) ws defined s the mitotic cell count per 10 high-power fields (HPFs), n re 0.183 mm 2. Two to 6 replicte sets of 10 HPFs were counted from 328 invsive brest crcinoms. Stndrd errors nd coefficients of vrition for men MI were compred with expected results predicted by the binomil distribution. Results. The boundries for MI tht seprted the dt into equl thirds (tertils) were 1.14 nd 5.33. Stndrd errors nd coefficients of vrition for MI followed distributions predicted by binomil probbility. Men coefficient of vrition ws 147% for the low tertil, 72% for the midtertil, nd 34.6% for the upper tertil. Conclusions. Stndrd errors for MI bsed on single count of 10 HPFs re too brod nd coefficients of vrition too lrge to be cceptble for clinicl use. This is explined s binomil probbility effect, possibly with contribution from tumor heterogeneity. Errors cn be reduced in proportion to the squre root of the number of sets of 10 HPFs counted. Tertil cutoffs of MI of the Nottinghm system currently used in brest crcinom grding re too high to be pplicble to the popultion we studied. We recommend vlidtion of cutoffs before they re pplied to prticulr popultion of brest crcinoms. Counting 5 sets of 10 HPFs is necessry to ccurtely rnk crcinoms with low MIs. (Arch Pthol Lb Med. 2009;133:1826 1833) Brest crcinom grding by histologic fetures hs been performed for more thn 50 yers. 1 Despite some technicl improvements, 2 grding, s commonly prcticed, hs not chieved sufficient reproducibility nd relibility to wrrnt its widespred cceptnce s bsis for clinicl decision-mking. 3 Its shortcomings include fuzzy definitions of grding criteri, indequte definition of optimum cutoffs, nd indequte smpling. The 3 grding components in common use re the degree of lumens or tubules (tubulrity) formed, nucler pleomorphism, nd mitotic figure count in 10 high-power microscopic fields (HPFs). We will refer to the men mitotic count per 10 HPFs when 1 or more sets of 10 HPFs re counted s the mitotic index (MI). Tubulrity nd mitotic count re potentilly quntifible on continuous scles, but the trditionl smple of 10 HPFs is exceedingly smll in reltion to the low mitotic frequency of mny crcinoms. Nucler pleomorphism is poorly defined feture, bsed on combintion of vribles tht re evluted subjectively. Mitotic count cutoffs re subject to lrge smpling errors, nd prognostic or predictive cutoffs re not well studied. 4 Accepted for publiction Jnury 8, 2009. From the Deprtment of Pthology, St. Luke s Hospitl, Chesterfield, Missouri (Dr Meyer); nd the Mchine Lerning Deprtment, NEC Lbortories Americ, Inc, Princeton, New Jersey (Drs Costto nd Grf). The uthors hve no relevnt finncil interest in the products or compnies described in this rticle. Reprints: John S. Meyer, MD, Deprtment of Pthology, St. Luke s Hospitl, 232 S Woods Mill Rod, Chesterfield, MO 63017-3417 (e-mil: john.meyer@stlukes-stl.com). Nevertheless, brest crcinom grding correltes with clinicl outcomes, nd it cn be performed inexpensively. Modern therpeutic dvnces in the tretment of brest crcinom require proper selection of ptients for n optiml effect. Newer methods of clssifiction, which include gene expression profiling, offer ttrctive biologic nd technicl rtionle. Genes regulting prolifertion re prominent in published expression profiles. 5 7 Technicl demnds nd expense currently mke this pproch inpplicble to lrge portion of the world s popultion. Therefore, we hve looked t wys of improving the grding system. Our gol ws to chieve sufficient reproducibility to mke the grding system useful in determining the need for djuvnt therpy nd for prediction of the response to different types of therpy. 8,9 Specific questions we sought to ddress included describing the coefficient of vrition (CV) of the mitotic index, determining cutoffs for the upper nd lower thirds of the distribution of the MI, nd optimizing the re of crcinom to be counted to blnce the ccurcy ginst the time expended. MATERIALS AND METHODS This work ws crried out during 4 yers within prctice of dignostic surgicl pthology (114 crcinoms) nd during review of deidentified specimens for the Coopertive Brest Cncer Tissue Resource 10 (214 crcinoms). Demogrphic dt were not vilble on some of the ltter ptients. All specimens of brest crcinom formlly exmined by one of us (J.S.M.) during the period were entered if t lest 2 sets of 10 HPFs could be counted. Of the 328 specimens, 26 (8%) were needle biopsy cores. Others were specimens removed by excisionl biopsy, tylectomy, or totl mstectomy. The ge of the ptient ws recorded in 227 instnces 1826 Arch Pthol Lb Med Vol 133, November 2009 Mesurement of Brest Cncer Mitotic Index Meyer et l
Tble 1. Type Neoplstic Cellulrity of Node-Negtive Brest Crcinoms Cses, No. Neoplstic Cells, No. Men/ Men/ 0.1 mm 2 HPF HPF, Medin HPF, Rnge Ductl 22 266 487 455 198 909 Lobulr 9 240 440 455 130 682 Tubulr 2 173 316 316 221 410 Mucinous 2 173 316 316 124 511 Medullry 2 292 534 534 401 670 Adenocystic 1 506 927 Ppillry 1 252 461 Abbrevition: HPF, high-power field. HPF re, 0.183 mm 2. (69%), with men of 59.3, medin of 60.0, nd extremes of 30 nd 89 yers. Axillry nodl sttus ws known in 187 cses (57%) by xillry dissection or sentinel node biopsy, nd of those, 20 (10.7%) showed metstsis. The low rte of nodl metstsis is ttributble to the preference of the investigtors who often requested node-negtive specimens from the tissue resource. Of the 310 specimens ccessioned consecutively, the frequency percentges for types of infiltrting crcinom were 79.4% ductl without further specifiction, 11.0% lobulr, 2.0% tubulr, 2.0% ppillry, 1.3% medullry, 1.3% mucinous, 0.7% denoid cystic, nd 2.9% other nd mixed types. In preliminry study, counts of neoplstic cells per unit re were done on specimens selected to represent severl types of brest crcinom (Tble 1). Highly cellulr regions of the vrious tumors were counted using n opticl grticule of 0.062-mm 2 re. From 2 to 17 (men, 14) 0.062-mm 2 fields were counted in 37 tumors. Mitotic counts were determined in more thn 1 nd up to 6 sets of 10 HPFs by single pthologist (J.S.M.) during 2004 to 2008. The res to be counted were selected from the most highly cellulr, well-preserved portions of the invsive neoplsm. No effort ws mde to determine the re with the highest mitotic count (the hot spot) before counting. To do so, would hve introduced sttisticl bis. The beginning field ws selected under scnning mgnifiction t which mitoses were not distinguishble. Thus, the first field counted might or might not contin mitotic figure. Fields were then counted systemticlly from left to right in descending rows using mnul mechnicl stge. Fields with poor cellulrity were skipped, nd the next good field ws counted. When neoplsm of comprbly good preservtion ws present in microsections from more thn 1 tissue block, severl blocks were included in the counts. The number of sets of 10 HPFs counted ws limited by the vilbility of tissue, prticulrly in core needle biopsies, or by time constrints. Criteri for recognition nd exclusion of mitosis followed those of Elston nd Ellis. 2 Counts of mitotic figures per 10 HPFs were recorded on continuous scle. A MI (men number of mitoses per 10 HPFs) ws clculted for ech specimen. Tertil cutoffs tht divided the dt set into equl thirds by mgnitude of the MI were then defined. We developed mthemtic model to which MI dt could be fit. To clculte the expected number of cells to be found in mitosis (x) within popultion of cells (n), given probbility of mitosis (p), nd probbility (q) of no mitosis, we used the following binomil distribution formul: n! x n x P{x} pq x!(n x)! Probbilities of certin numbers of mitotic figures being found in 10 HPFs could then be clculted for vrious levels of cellulrity. We clculted results for 100 nd for 200 cells per 0.062- mm 2 field, equivlent to 2950 nd 5900 cells/10 HPFs. Those levels fell within the rnge of cellulrity we observed (Tble 1). Tbles 2 nd 3 show predicted dispersions of mitotic counts per 10 HPFs, ccording to the binomil distribution for low nd high levels of cellulrity, nd rnge of probbilities tht given cell will be in mitosis. For exmple, for tumors with low cellulrity (Tble 2, row 2), probbility of mitosis per cell of 0.0002 predicts probbility of finding no mitosis in 10 HPFs of 0.5543. Reding further in row 2, the probbility of finding 1 mitosis is 0.3271; for 2, it is 0.0965; for 3, it is 0.0190; nd for 5, it is 0.0003. If cellulrity is high nd probbility is still 0.0002, probbilities of finding certin numbers of mitoses re shifted towrd higher numbers (Tble 3, row 2). If the probbility per cell is set t 0.005, nd cellulrity is 5900/10 HPFs, the probbilities of finding 10 or fewer mitotic figures per 10 HPFs will essentilly be 0 (Tble 3, third row from bottom). For exctly 20, 30, nd 50 mitotic figures, the corresponding probbilities re 0.0157, 0.0727, nd.0001, respectively. Probbilities for intervening nd lrger numbers of mitotic figures re not shown, but they would dd up to 100%. To test how well the mesured dt fit the binomil model, we clculted the theoretic 95% confidence intervls. These re the intervls within which 95% of ll observtions re expected. Tble 2. Probbilities (P ) of Discrete Numbers of Mitotic Figures Per 10 High-Power Fields (HPFs), With 2950 Cells/ 10 HPF P {mitosis/cell} b Men/ 10 HPFs c P {0} d P {1} P {2} P {3} P {5} P {10} P {20} P {30} P {50} 0.0001 0.295 0.7445 0.2197 0.0324 0.0032 0.001 0 0 0 0 0.0002 0.590 0.5543 0.3271 0.0965 0.0190 0.0003 0 0 0 0 0.0003 0.885 0.4127 0.3653 0.1616 0.0477 0.0019 0 0 0 0 0.0004 1.180 0.3072 0.3626 0.2140 0.0841 0.0058 0 0 0 0 0.0005 1.475 0.2287 0.3375 0.2489 0.1224 0.0133 0 0 0 0 0.0006 1.77 0.1702 0.3015 0.2669 0.1575 0.0246 0 0 0 0 0.0010 2.95 0.0523 0.1543 0.2278 0.2241 0.0975 0.0007 0 0 0 0.0015 4.425 0.0119 0.0529 0.1172 0.1729 0.1694 0.0095 0 0 0 0.002 5.90 0.0027 0.0161 0.0476 0.0937 0.1633 0.0385 0 0 0 0.003 8.85 0.0001 0.0013 0.0056 0.0165 0.0648 0.1166 0.0005 0 0 0.004 11.80 0 0.0001 0.0005 0.0020 0.0142 0.1084 0.0084 0 0 0.005 14.75 0 0 0 0.0002 0.0023 0.0526 0.0383 0.0002 0 0.01 29.50 0 0 0 0 0 0 0.0156 0.0727 0.0001 0.02 59.04 0 0 0 0 0 0 0 0 0.0272 HPF re, 0.183 mm 2 ; mgnifiction, 400. Entries in the tble re theoretic binomil probbilities of observing x number of mitotic figures. b Probbility tht ny 1 cell will be in mitosis. c Men number of mitotic figures per 10 HPF. d Probbility of counting no mitotic figure in set of 10 HPFs selected t rndom. Becuse of spce limittions, probbilities for only selected numbers of mitotic figures re shown. Arch Pthol Lb Med Vol 133, November 2009 Mesurement of Brest Cncer Mitotic Index Meyer et l 1827
Tble 3. Probbilities (P) of Discrete Numbers of Mitotic Figures per 10 High-Power Fields (HPFs) With 5900 Cells/ 10 HPF P {mitosis/cell} b Men/ 10 HPFs c P {0} d P {1} P {2} P {3} P {5} P {10} P {20} P {30} P {50} 0.0001 0.59 0.5543 0.3271 0.0965 0.0190 0.0003 0 0 0 0 0.0002 1.18 0.3072 0.3626 0.2140 0.0841 0.0059 0 0 0 0 0.0003 1.77 0.1703 0.3015 0.2669 0.1574 0.0246 0 0 0 0 0.0004 2.36 0.0944 0.2228 0.2630 0.2069 0.0576 0.0001 0 0 0 0.0005 2.95 0.0523 0.1544 0.2278 0.2240 0.0975 0.0007 0 0 0 0.0006 3.54 0.0290 0.1027 0.1818 0.2146 0.1344 0.0025 0 0 0 0.0010 5.90 0.0027 0.0161 0.0476 0.0937 0.1633 0.0386 0 0 0 0.0015 8.85 0.0001 0.0013 0.0056 0.0165 0.0648 0.1166 0.0005 0 0 0.002 11.80 0 0.0001 0.0005 0.0020 0.0143 0.1084 0.0084 0 0 0.003 17.70 0 0 0 0 0.0003 0.0170 0.0771 0.0021 0 0.004 23.60 0 0 0 0 0 0.0008 0.0665 0.0327 0 0.005 29.50 0 0 0 0 0 0 0.0157 0.0727 0.0001 0.01 58.41 0 0 0 0 0 0 0 0 0.0272 0.02 118.1 0 0 0 0 0 0 0 0 0 HPF re, 0.183 mm 2 ; mgnifiction, 400. Entries in the tble re theoretic binomil probbilities of observing x number of mitotic figures. b Probbility tht ny 1 cell will be in mitosis. c Men number of mitotic figures per 10 HPF. d Probbility of counting no mitotic figure in set of 10 HPFs selected t rndom. Becuse of spce limittions, probbilities for only selected numbers of mitotic figures re shown. Mny vlues outside this rnge would put into question whether the model is pproprite. In Tble 4, the theoretic 95% intervls re shown for few expected MIs. These vlues were clculted by interpolting the binomil curves with third-order polynomils nd numeric integrtion of the re under these curves. The re below the lower mrk corresponds to 2.5% probbility; similrly, the re bove the higher mrk lso corresponds to 2.5% probbility, nd 95% of ll cses re expected to fll within the re between the 2 mrks. Note the degree of improvement in predicted ccurcy s the number of sets of 10 HPFs tht re counted increses from 1 (column 4) to 5 (column 7). Probbilities for mitoses per 10 HPFs cn lso be estimted by the norml pproximtion to the binomil s expressed by the following formul: x x x (np) Z npq This provides quick nd esy wy of checking experimentl results nd is resonbly ccurte when np 5. The Z reltes to the probbility (P) of chieving x number of mitotic figures; q 1 p; ndz 1 is equivlent to 1 stndrd devition of the men. When Z 1.96, the probbility of chieving higher number of mitotic figures is less thn 0.05. By symmetry, when Z 1.96 stndrd devitions ( 2 ), the probbility of fewer mitotic figures is less thn 0.05. When Z 0, the probbility of chieving mitotic figure on ny tril is np. The norml pproximtion of the binomil ws used to clculte probbilities for distributions of MI cross the tertil cutoff boundries. Results were closely similr to those clculted by numeric, interpolted probbility distribution for n MI equl to or lrger thn 6, but they becme incresingly divergent for smller MIs. The CV of n MI is defined s the stndrd devition divided by the men times 100%. Frequency distributions, sctter plots, curve fitting, nd other sttisticl nlyses were performed on computers running JMP 6 sttisticl softwre (SAS Institute Inc, Cry, North Crolin) nd NEC MiLDe proprietry softwre (NEC Lbortories, Inc, Cupertino, Cliforni). Significnce tests were 2-tiled. The project ws conducted with pprovl of the St. Luke s Hospitl Institutionl Review Bord for Humn Subjects Reserch. RESULTS Neoplstic cellulrity did not vry gretly between infiltrting ductl nd infiltrting lobulr crcinom in our limited survey (Tble 1). A wider vrition ws seen between individul crcinoms of given type. The few other crcinoms of specil types studied fell within the rnges of the ductl nd lobulr crcinoms, except for lowgrde denocystic crcinom composed of smll cells pcked closely together with low MI. Cellulrity ws not correlted with MI in 37 specimens with both mesure- Tble 4. Predicted Coefficients of Vrition (CV) nd 95% Confidence Limits (CL) for Mitotic Figures per 10 High-Power Fields (HPFs) for n Sets of 10 HPFs With 5900 Cells/10 HPFs Men/ SD 10 HPFs b (n 1) CV, % (n 1) 95% CL c (n 1) 95% CL (n 2) 95% CL (n 3) 95% CL (n 5) 1 1.00 100.0 0.0 3.4 0.1 2.6 0.1 2.3 0.2 1.9 2 1.41 70.7 0.1 5.3 0.3 4.2 0.5 3.7 0.8 3.3 3 1.73 57.7 0.3 6.9 0.8 5.6 1.2 5.1 1.5 4.5 4 2.00 50.0 0.7 8.3 1.5 6.9 1.9 6.3 2.4 5.8 5 2.23 44.7 1.2 9.8 2.2 8.3 2.6 7.6 3.1 7.0 6 2.45 40.8 1.7 11.2 2.8 9.6 3.4 8.8 3.9 8.1 7 2.65 37.8 2.3 12.9 3.5 10.9 4.2 10.1 4.7 9.4 8 2.83 35.3 2.9 14.0 4.3 12.0 4.9 11.3 5.6 10.5 9 3.00 33.3 3.5 15.4 5.0 13.2 5.8 12.5 6.4 11.6 10 3.16 31.6 4.3 16.6 5.9 14.5 6.5 13.7 7.2 12.7 HPF re, 0.183 mm 2 ; mgnifiction, 400. Entries in the tble re theoretic binomil probbilities for the mgnitude of CV. b Refer to Tble 3 for the probbility of mitosis in ny given cell. c The CLs were determined by numeric integrtion of interpolted probbility distribution. 1828 Arch Pthol Lb Med Vol 133, November 2009 Mesurement of Brest Cncer Mitotic Index Meyer et l
Figure 1. Coefficient of vrition (CV) s function of men mitotic count per 10 highpower fields (HPFs; mitotic index [MI]) for the complete dt set. A third-order polynomil regression is shown in blck. The thick gry line shows the vlues predicted by the binomil distribution (see Figure 5 for clcultions). ments (R 2 0.023, P.37 by nlysis of vrince). At lest 10 of the 0.062-mm 2 fields were counted in 19 tumors. The CV of these counts rnged from 14% to 55%, with men of 33%. The re of n HPF is 0.183 mm 2, 3 times lrger thn the field used for cell counts. Dividing by the squre root of 3 gives prediction of the men CV for replicte cellulrity counts on HPFs of 19%. A centrl observtion of our study is close fit of the distributions of mitotic counts to the binomil model. Stndrd errors nd CVs of the MI mesurements re lrge nd follow closely the mgnitudes predicted by the binomil model (Figure 1). This indictes poor reproducibility of mitotic counts per 10 HPFs for low MIs. Another importnt finding is tht the MI is distributed lognormlly, with strong negtive skew (Figure 2), resulting in tertil cutoffs incomptible with those of the Nottinghm system. More thn 1 mitotic count of 10 HPFs ws obtined on 328 infiltrting brest crcinoms. Two replicte counts were obtined from 70 specimens (21%), 3 replictes from 154 specimens (47%), 4 replictes from 64 specimens (20%), 5 replictes from 18 specimens (5%), nd 6 replictes from 22 specimens (7%). Counts of zero mitosis were observed 16 times in 2 sets of 10 HPFs, 20 times in 3 sets, twice in 4 sets, nd twice in 5 sets. In Tble 3, the probbility of mitotic figure of 0.0001 in ech cell with 590 cells/hpf would equte to men count of 0.59 mitotic figures per 10 HPFs. The ctul counts, therefore, support the existence of mitotic rtes t or less thn the 0.0001 probbility level per cell. The highest mitotic counts we observed were 48 nd 78 per 10 HPFs. Such high counts correspond to probbilities of mitosis of 0.008 to 0.013 per cell if the tumor is richly cellulr (5900 cells/10 HPFs) or 0.016 to 0.026 if it is poorly cellulr (2950 cells/10 HPFs). The probbility of finding brest crcinom cell in mitosis cn vry by fctor of 100 to 200, depending on the prolifertion rte of the prticulr crcinom. When theoretic vlues for the CVs of MIs in Tble 4 re compred with observed vlues in Tble 5, the distributions shown in the 2 tbles re seen to be similr. Both show strong decline of CVs with incresing MIs. Pre- Figure 2. Frequency distribution of men mitotic index (MI) for the consecutive dt set. Eighteen specimens selected to enrich the set with highly prolifertive crcinoms were omitted. The lognorml nture of this distribution is not evident here becuse numerous specimens with low or no MI re included in the fr left br. Arch Pthol Lb Med Vol 133, November 2009 Mesurement of Brest Cncer Mitotic Index Meyer et l 1829
Tble 5. Actul Coefficients of Vrition (CV) of Mitotic Figures per 10 High-Powered Fields (HPFs; Mitotic Index [MI]) Mesured in 270 Infiltrting Brest Crcinoms for Which 2 6 Sets of 10 HPFs Were Counted Tertil 1 2 3 MI intervl defining tertil b 0 1.14 1.15 5.32 5.33 78 Specimens, No. 63 c 103 104 Men CV, % 147 72.0 34.6 Medin CV, % 141 70.5 31.4 HPF re, 0.183 mm 2 ; mgnifiction, 400X; CV entries in the tble re mesured mgnitudes. b Rnge of men MI/10 HPFs defining the tertil. c The CVs could not be clculted for 40 specimens with men MI/ 10 HPF of 0. dicted nd observed CVs for the lower tertil of MIs rech 100% or more for single set of 10 HPFs. The rnge of CVs predicted for the high tertil is 43% (for MI 5.33) to 13.8% (for MI 78), which compres with the observed men CV for the tertil of 34.6%. In the complete dt set, in which consecutive ccessions were supplemented by tumors selected for high MI, liner reltionship of the stndrd error of the men to the men MI ws demonstrted (dt not shown). A thirdorder regression of the CV on the MI ws fit to the mesured dt, nd theoretic vlues of the CVs, s predicted by the binomil distribution, showed similr fit (Figure 1). The mesured CV is consistently close to the CV predicted by the mthemtic model. We used the consecutive dt set to evlute results of the clssifiction with different MI cutoffs. Cutoffs of pproximtely 0 6, 7 12, nd greter thn 12 mitoses per 10 HPFs with field res of 1.64 to 1.87 mm 2 hve been proposed by the Nottinghm group 2 to seprte low from mid nd mid from high. Our HPF re of 0.183 mm 2 fell within this intervl. When multiple sets of 10 HPFs re counted, one result cn fll to the left nd nother to the Figure 3. Frequency of overlp of tertil cutoffs for mitotic indices per 10 high-power fields (HPFs) when multiple sets of 10 HPFs were counted in the consecutive dt set. Cutoffs of 1.14 nd 5.35 (intervls, 0 1.14, 1.15 5.35, 5.36 high), s defined in the current study, were used. Percentges re stted t tops of columns. Abbrevitions t the bottoms of columns: N, no overlp; YL, overlp of low cutoff; YH, overlp of high tertil cutoff; YLH, overlp of both cutoffs. Figure 4. Frequency of overlp of tertil cutoffs for mitotic indices per 10 high-power fields (HPFs) when multiple sets of 10 HPFs were counted in the consecutive dt set. Cutoffs of 6.9 nd 12.9 (intervls, 0 6.9, 7.0 12.9, nd 13 high), s defined in Nottinghm study, were used. Percentges re stted t tops of columns. Abbrevitions t the bottoms of columns: N, no overlp; YL, overlp of low cutoff; YH, overlp of high tertil cutoff; YLH, overlp of both cutoffs. right of cutoff. The proportion of specimens with counts tht overlp cutoff reltes to the probbility tht smple from given tumor will be ssigned to the sme tertil s its true popultion men. The likelihood of n incorrect tertil ssignment increses in proportion to the mgnitude of the CVs of the MI nd the proximity of the popultion men to the cutoff. We evluted the frequency of overlp of our tertil cutoffs (Figure 3) in comprison with the Nottinghm cutoffs (Figure 4) when 2 to 6 sets of HPFs were counted. Overlps occurred more frequently with the lower cutoffs. This is n expected result of the strong positive skew of the MI frequency distribution. More thn two-thirds of the MIs were less thn the low Nottinghm cutoff. The frequency of cutoff overlp incresed from 31.3%, counting both low nd high tertil boundry overlps, for the 67 specimens with only 2 sets of HPFs counted, to 41.3%, for the 152 specimens with 3 sets counted, nd to 50.0%, for the 60 specimens with 4 sets counted. Cutoff overlp ws seen in 48.4% of 31 specimens with 5 or 6 sets counted. Lower overlp rtes when few sets re counted cn be ttributed to the decresed likelihood of obtining extreme vlues when fewer replicte counts re done. Morphologic evidence of heterogeneity within brest crcinoms ws noted in 6 specimens, but heterogeneity did not seem to ffect CVs. Coefficients of vrition for the MI mens for the 6 specimens were respectively 49% t 14.5, 45% t 7.33, 17% t 6.67, 95% t 6.25, 47% t 5.33, nd 96% t 3.33. These CVs re close to the CV mens for the MI ctegories shown in Tble 5. To investigte the possibility of systemtic bis relted to ftigue or some other cuse, we compred MI counts of the whole dt set ccording to the order in which they were performed. This demonstrted slightly incresed men MIs for the second count versus the first count (Tble 6). The men of counts for the first set were slightly lower thn for subsequent sets. The difference ws significnt by pired t test for set 1 versus set 3, nd by nlysis of vrince for sets 1, 2, nd 3. Other differences were not significnt. This seems counterintuitive becuse our pro- 1830 Arch Pthol Lb Med Vol 133, November 2009 Mesurement of Brest Cncer Mitotic Index Meyer et l
Figure 5. Clcultions of CV vlues predicted by the binomil distribution (gry line in Figure 1). cedure clled for selection of the most vigorous-ppering portion of the tumor for the first count. COMMENT Low-prolifertion indices predict low rtes of erly clinicl progression. 11 15 High rtes of prolifertion predict erly relpse nd high probbility of histologic complete regression with neodjuvnt cytotoxic therpy. 8,9 Mitosis is the definitive event tht completes the cell cycle. It reltes unequivoclly to the genertion of new cells. Of vrious indirect mesurements of cell-prolifertion, mitosis hs the shortest durtion nd, therefore, yields the lowest counts. Ki-67 growth frctions in brest crcinoms rnge from 90% to 1%, S-phse frctions from 40% to less thn 0.1%, nd mitotic counts per 10 HPFs from 2% to less thn 0.001%. Ten HPFs of cellulr res of brest crcinoms usully contin 1200 to 9000 cells. One-third of brest crcinoms in our survey hd no more thn 1.14 mitoses/10 HPFs. The 95% confidence limits of this cutoff extend from 0 to 3.49 mitotic figures per 10 HPFs if only 1 set of 10 HPFs is counted. Therefore, we conclude tht count of 10 HPFs is indequte. Cn this result be improved? The stndrd devition nd stndrd error of the men re reduced in proportion to the squre root of the number of sets counted. A modest but meningful improvement in confidence limits is chieved by counting 2 or 3 sets of HPFs. Estimtes pproching the precision of most clinicl lbortory tests would require counting 10 or more. Coefficients of vrition for most nlytes mesured in the clinicl chemistry lbortory re less thn 5%, some even less thn 1%. Coefficients of vrition for hemoglobin nd red blood cell counts re nerly 5%, nd those for white blood cell counts nd pltelets re up to 15% by utomted techniques. The low frequency of mitoses t the criticl lower one-third cutoff of 1 or 2 mitoses/10 HPFs is serious hndicp to precision. Our cutoffs for the lower nd middle one-third distributions of mitotic figures per 10 HPFs re distinctly lower thn those now generlly ccepted. 2,16 This my be due to, in prt, the smller sizes of invsive brest crcinom in the er of extensive popultion screening, when compred with prior yers. To investigte whether the pthologist (J.S.M.) who counted mitotic figures in the current study might hve prticulrly high threshold for recognizing them, we exmined 2 udits of Coopertive Brest Crcinom Tissue Resource specimens. In these udits, pthologists ssigned grdes of 1, 2, or 3 to numbers of mitotic figures per 10 HPFs ccording to Nottinghm criteri. 4 J.S.M. rnked highest mong 7 pthologists in MI counts for 18 crcinoms with men score of 1.4 (rnge, 1.2 1.4; yer 2005), nd fourth mong 8 pthologists for nother 16 crcinoms (men score, 1.6; rnge, 1.5 1.7; yer 2006). Tertil cutoffs of the current study were similr to those for set of more thn 600 node-negtive invsive brest crcinoms in which most mitotic counts were performed by pthologists other thn J.S.M. 4 Brest crcinoms with prolifertion indices in the low tertil hd distinctly better prognosis thn crcinoms in the mid tertil in 2 sets of ptients we hve studied, 4,11 nd these findings re consistent with reports by other investigtors. 13 15 Two prior studies, one of stge 1 brest crcinoms, 17 the other of lymph node positive crcinoms, 18 ech found low cutoff of 2 mitotic figures per 10 HPFs Tble 6. Comprison of First, Second, Third, nd Fourth Sets of Mitotic Counts Set, No. Specimens, No. Men Mitoses/10 HPFs Pired t Test Significnce (P )by Correltion Coefficient 1 vs 2 328 6.24 vs 6.91.03.88 1 vs 3 258 6.06 vs 6.92.008.885 1 vs 4 105 7.27 vs 7.79.52.77 1 vs 5 41 6.78 vs 8.54.12.91 2 vs 3 258 6.83 vs 6.92.81.865 2 vs 4 105 8.41 vs 7.79.50.73 3 vs 4 105 8.07 vs 7.79.76.74 Abbrevition: HPFs, high-powered fields. HPF re, 0.183 mm 2 ; mgnifiction, 400. Arch Pthol Lb Med Vol 133, November 2009 Mesurement of Brest Cncer Mitotic Index Meyer et l 1831
ws closer to the boundry of the low tertil nd betterdefined good outcome thn the Nottinghm cutoffs. The current Nottinghm cutoffs group low tertil nd mid tertil ptients together. To seprte the 2 prognosticlly different groups, the cutoffs must be revised downwrd. Tertil cutoffs my not be trnsferble from one brest crcinom popultion to nother. Age, tumor size, nd lymph nodl sttus correlte with prolifertion indices, 11 13 nd these dt were not vilble for 83 of our specimens. For the 227 specimens with dt, tertil cutoffs were 1.00 nd 5.03. For the 83 without dt, cutoffs were slightly higher, 1.33 nd 5.35, but still much lower thn the Nottinghm cutoffs. Besides smpling effects following the binomil distribution, rel regionl differences in prolifertion indices cn exist within brest cncers. They cn be cused by vritions in blood circultion or by presence of heterogeneous clones within the tumor. 19,20 In prior study, 21 using tritited thymidine lbeling to identify S-phse cells, evidence of regionl vrition within brest crcinoms ws detected in 61% of 50 tumors. The vritions were modest, with rtios of high to low counts of 2:1 to 3:1. The men thymidine lbeling index of those invsive brest crcinoms ws 7.4%. This compres with men probbility of mitosis of 0.1% to 0.2% for our consecutive set of 310 crcinoms. Approximtely 50 cells re lbeled with tritited thymidine or its nlogue, bromodeoxyuridine, for ech cell in mitosis. For this reson, DNA precursor uptke is much more sensitive for detection of heterogeneity in prolifertion rte thn MI. We noted morphologic heterogeneity ccompnied by suggestion of disprity in the MI within 5 of 328 crcinoms (1.5%), but sttisticlly significnt regionl differences in MIs were not demonstrted. Differences in thresholds for recognition of mitotic figures my explin, in prt, the low levels of greement often noted mong pthologists who count the sme specimen. 22 The results we hve presented suggest to us tht rndom vrition in the distribution of mitotic figures is much more importnt cuse of disgreement, prticulrly for crcinoms with low MI. Our observed error profile ws closely predicted by the binomil distribution, independent of ny fctor representing vrition in recognition of mitotic figures by the microscopist. The binomil distribution suffices to explin the lrge CVs we observed. Nevertheless, vritions relted to criteri for recognition of mitotic figures my ssume n incresed importnce when the smpling error is reduced by counting dded sets of HPFs. Surrogte mrkers for mitotic figures, prticulrly those tht increse the count by mrking cells committed to mitosis but not necessrily in mitosis, give promise of utility. 16 If fixed probbility of mitosis is postulted, differences in popultion density of mlignnt cells cn predictbly influence the MI. These differences cn be up to 8-fold from one tumor to nother, when the more-cellulr res re compred, nd up to n 8-fold difference in MI between tumors could then result, even though the underlying probbilities of mitosis were the sme. We re interested not in MI differences between tumors, but in the ccurcy of mesuring the MI in ny prticulr tumor. Here, only within-tumor vribility in the cellulrity is importnt, nd this is equivlent to reltively smll CV of 19%. The need for counting more thn 1 set of 10 HPFs hs been stted by Nottinghm uthors, 2 but it hs not been emphsized sufficiently. Mitotic indices derived from single count of 10 HPFs re not relible in theory or in prctice. A single count my be bised by unforeseen resons. The significntly reduced result for our first count might result from wrm-up effect, wherein recognition of mitoses is improved with counting. Mitotic figures my vry in ppernce from one specimen to nother in subtle wys, determined by the chromosome nd spindle numbers. These differences could render mitotic figures in given tumor more redily identifible fter the observer dpts to the predominnt pttern present. However the most importnt determinnt of ccurcy, prticulrly in the low rnge of the MIs, is the error introduced by the binomil distribution of mitotic figures. This cn be ddressed only by incresing the smple size. The optimum number of sets of 10 HPFs to be counted depends on severl considertions. Accurte rnking of crcinoms requires counting incresing numbers of sets of 10 HPFs s the MI decreses. Our results suggest tht 3 or more sets of 10 HPFs should be counted when the first count is less thn 6, nd tht 5 sets re needed to clssify crcinom s lower tertil versus mid tertil when the count is ner the cutoff. A single count showing 6 or more mitotic figures per 10 HPFs ppers to be sufficient to exclude popultion-men MI of 1 per 10 HPFs or less, but more sets should be counted if the gol is to differentite mid tertil from top tertil. We propose tht pthologists or technologists engged in brest crcinom grding tbulte the results of their mitotic counts of 25 consecutive, unselected, node-negtive tumors nd 25 consecutive, unselected, node-positive tumors. This will estblish personl benchmrks for men nd medin MIs nd lower one-third nd upper one-third cutoffs. Similr qulity control mesures hve long been prcticed in other sections of clinicl lbortories nd re necessry if MI is to become cliniclly relible, reimbursble ssy. Mitotic index results should be reported on continuous scle in ddition to tertil results. Even if up to 5 sets of 10 HPFs re counted, CVs will not be sufficiently smll to ensure consistent ssignment to first tertil versus second tertil for tumors with MI vlues ner the cutoff. References 1. Bloom HJG, Richrdson WW. Histologicl grding nd prognosis in brest cncer. Br J Cncer. 1957;11(3):359 377. 2. Elston CW, Ellis IO. Pthologicl prognostic fctors in brest cncer, 1: the vlue of histologicl grde in brest cncer: experience from lrge study with long-term follow-up. Histopthology. 1991;19(5):403 410. 3. Consensus Pnel. 1997 Updte of recommendtions for the use of tumor mrkers in brest nd colorectl cncer. J Clin Oncol. 1998;16(2):793 795. 4. Meyer JS, Alvrez CA, Milikowski C, et l. Brest crcinom mlignncy grding by Bloom-Richrdson system vs prolifertion index: reproducibility of grde nd dvntges of prolifertion index. Mod Pthol. 2005;18(8):1067 1078. 5. vn t Veer LJ, Di H, vn de Vijver MJ, et l. Gene expression profiling predicts clinicl outcome of brest cncer. Nture. 2002;415(6871):530 536. 6. vn de Vijver JM, He UD, vn t Veer LJ, et l. A gene-expression signture s predictor of survivl in brest cncer. N Engl J Med. 2002;347(25):1999 2009. 7. Pik S, Shk S, Tng G, et l. A multigene ssy to predict recurrence of tmoxifen-treted, node-negtive brest cncer. N Engl J Med. 2004;351(27): 2817 2826. 8. Amdori D, Nnni O, Mrgolo M, et l. Disese-free survivl dvntge of djuvnt cyclophosphmide, methotrexte, nd fluorourcil in ptients with node-negtive, rpidly proliferting brest cncer: rndomized multicenter study. J Clin Oncol. 2000;18(17):3125 3134. 9. Andre F, Khlil A, Slimne K, et l. Mitotic index nd benefit of djuvnt nthrcycline-bsed chemotherpy in ptients with erly brest cncer. J Clin Oncol. 2005;23(13):2996 3000. 10. Glss AG, Donis-Keller H, Mies C, et l. The Coopertive Brest Cncer 1832 Arch Pthol Lb Med Vol 133, November 2009 Mesurement of Brest Cncer Mitotic Index Meyer et l
Tissue Resource: rchivl tissue for the investigtion of tumor mrkers. Clin Cncer Res. 2001;7(7):1843 1849. 11. Meyer JS, Province M. Prolifertive index of brest crcinom by thymidine lbeling: prognostic power independent of stge, estrogen nd progesterone receptors. Brest Cncer Res Tret. 1988;12(2):191 204. 12. Silvestrini R, Didone MG, Luisi A, et l. Cell prolifertion in 3,800 nodenegtive brest cncers: consistency over time of biologicl nd clinicl informtion provided by 3 H-thymidine lbelling index. Int J Cncer. 1997;74(1):122 127. 13. Goodson WH III, Moore DH II, Ljung B-M, et l. The prognostic vlue of prolifertion indices: study with in vivo bromodeoxyuridine nd Ki-67. Lbeling index s prognostic mrker in brest cncer. Brest Cncer Res Tret. 2000; 59(2):113 123. 14. Pinder SE, Wencyk P, Sibbering DM, et l. Assessment of the new prolifertion mrker MIB1 in brest crcinom using imge nlysis: ssocitions with other prognostic fctors nd survivl. Br J Cncer. 1995;71(1):146 149. 15. Rudolph PMG, McGrogn G, Bonichon F, et l. Prognostic significnce of Ki-67 nd topoisomerse II expression in infiltrting ductl crcinom of the brest: multivrite nlysis of 863 cses. Brest Cncer Res Tret. 1999;55(1): 61 71. 16. Sklnd I, Jnssen EAMK, Gudlugsson E, et l. Phosphohistone H3 expression hs much stronger prognostic vlue thn clssicl prognostictors in invsive lymph node-negtive brest cncer ptients less thn 55 yers of ge. Mod Pthol. 2007;20(12):1307 1315. 17. Frkovic-Grzio S, Brocko M. Long term prognostic vlue of Nottinghm histologicl grde nd its components in erly (pt1n0m0) brest crcinom. J Clin Pthol. 2002;55(2):88 92. 18. Simpson JKF, Gry R, Dressler LG, et l. Prognostic vlue of histologic grde nd prolifertive ctivity in xillry node-positive brest cncer: results from the Estern Coopertive Oncology Group Compnion study, EST 4189. J Clin Oncol. 2000;18(10):2059 2069. 19. Nowell PC. The clonl evolution of tumor cell popultions. Science. 1976; 194(4260):23 28. 20. Fidler IJ, Hrt IR. The origin of metsttic heterogeneity in tumors. Eur J Cncer. 1981;17(5):487 494. 21. Meyer JS, Wittliff JL. Regionl heterogeneity in brest crcinom: thymidine lbelling index, steroid hormone receptors, DNA ploidy. Int J Cncer. 1991; 47(2):213 220. 22. Chowdhry N, Mukth RP, Lobo FD, Kini H, Vrghese R. Interobserver vrition in brest cncer grding: sttisticl modeling pproch. Anl Qunt Cytol Histol. 2006;28(4):213 218. Arch Pthol Lb Med Vol 133, November 2009 Mesurement of Brest Cncer Mitotic Index Meyer et l 1833