Prediction of Survival in Adults with Acquired Bi- or Pancytopenia

Scand I Haematol(l984) 32,119-129

Prediction of Survival in Adults with Acquired Bi- or Pancytopenia

L. CLEMMESEN,l M. M0RK HANSEN,' L. MORTENSEN,3 T. THIEDE2 & AA. VIDEBRK'

University Departments of I Medicine and Haematology C and Pathology, Gentofte Hospital and

University Department for Data Processing, Herlev Hospital, Copenhagen, Denmark

In a retrospective study of 44 adults with acquired bi- or pancytopenia without evidence of any causal disorder, the survival curve suggested the existence of a subgroup of short survivors, mainly with aplastic anaemia, with death within 4 months. The initial values of 14 single clinical, blood and bone marrow variables were significantly associated with survival < 4 months. Stepwise multiple logistic regression analyses identified 2 combinations of variables displaying significant simultaneous associations with short survival: (i) increased % of non-myeloid bone marrow cells and haemorrhagic manifestations initially; (ii) increased % of non-myeloid marrow cells, circulating erythroblasts and no history of any drug exposure. The predictive capacities of a resulting estimate of the probability of short survival and of previously introduced prognostic indices were approximately equal. The frequency of a correct prediction of a survival shorter than 4 months was in the range of 0.71-0.78 and that of longer survival in the range of 0.74-0.94.

Key words: aplastic anaemia - multivariate statistical analysis - pancytopenia - prognostic factors - prognostic index

Accepted for publication July 30, 1983

Correspondence to: Dr. Lars Clemmesen, Ved Hojmosen 111, DK-2970 Horsholm, Denmark

ABBREVIATIONS A A = aplastic anaemia B/PC = bicytopenia or pancytopenia DS(Cam) = degree of severity of Camitta et a1 (1) EP = estimated probability of survival

< 4 months (this study) EP(Roz) = estimated probability of short survival

of Rozman et a1 (4) LT = value of log likelihood ratio test

statistic PI = prognostic index (this study)

PI(Lyn) PI(Naj)

W2T

= prognostic index of Lynch et a1 (2) = prognostic index of Najean & Pecking

= value of Wilcoxon two-sample test (3)

statistic

The survival of patients with aplastic anaemia (AA) ranges from days to years. In a recent study the survival curve suggested

120 CLEMMESEN ET AL

the existence of subgroups characterized by survival shorter and longer than 4 months

Early identification of patients with a poor prognosis would be advantageous at the selection of patients for bone marrow transplantation or at the stratification of patients in therapeutical trials. For this application Camitta et a1 (1) introduced an arbitrarily defined degree of severity (DS(Cam)), while Lynch et a1 (2) and Najean & Pecking (3) both devised prognostic indices (PI(Lyn) and PI(Naj)), by means of discriminant analysis. Ultimately, Rozman et a1 (4) presented an estimated probability of early death (EP(Roz)) derived by linear logistic regression analysis.

This paper presents a retrospective investigation of patients with bicytopenia or pancytopenia (B/PC), i.e. with at least 2 of the following: anaemia, granulocytopenia, and thrombocytopenia, and no evidence of any causal disorder. This syndrome is al- most always present at first admission and includes patients with AA. The aim of the study was (i) to see whether the existence of a subgroup of short survivors could be confirmed; if so, (ii) to identify single or combined variables predictive of short survival; and (iii) to evaluate the precision of the prognostic criteria and indices previously suggested for prediction of short survival.

(2) *

PATIENTS AND METHODS

The selection of patients and the methods employed have been described previously (5), and only the main points will be dealt with.

Patients 44 patients, 21 men and 23 women, median age 55.5 years (range 13-83), admitted to Medical Department C, Gentofte University Hospital, during the period

April 1962-April 1974 fulfilled the criteria: B/PC not secondary to any recognizable disorder, no history of familial occurrence, and no cytostatic or actinic treatment. Anaemia, granulocytopenia, and thrombocytopenia were defined as Hb < 117 g/l in women, < 132 g/l in men, granulocyte count < 1.8 x 109/1 and thrombocyte count < 150 x 109/l.

The primary clinical diagnoses were AA (8 patients), pancytopenia (14), thrombocytopenia (4), agranulo- cytosis (4), various anaemia diagnoses (lo), and other diagnoses (4). Retrospectively, the patients were cate- gorized as AA (23 patients, of which 1 developed acute leukaemia terminally), preleukaemia (Il), agranulo- cytosis accompanied by anaemia or thrombocytopenia (4), and B/PC with normo- or hypercellular bone marrow (6) .

The period of observation commenced when a suc- cessful bone marrow specimen was obtained. The criteria of entrance were fulfilled in all patients except 2 at the first admission to hospital. 1 patient had a preceding phase of anaemia only, lasting 4 months, the other entered 2 years after recognition of bicytopenia. All patients could be traced to either death or follow-up at Oct 1975. All 34 deceased patients except 5 had a post-mortem examination.

The treatment after diagnosis was individualized according to the degree of cytopenia and the occurrence of complications.

Variables The values of the variables at the start of observation were obtained from the actual case records and through blind revision of smears of blood and bone marrow. The revision included case records and smears from possible previous admissions to other hospitals. The absolute cellularity of the bone marrow and its various compartments were estimated as polytomous (multi- classed qualitative) variables, in 15 patients from needle or surgical biopsies, in 28 patients from sections of marrow clots, while in 1 patient only a smear of marrow aspirate was available.

A history of exposure to drugs or other chemicals was accepted, if the exposure had occurred within 1 year prior to the first admission, but before onset of symptoms which could be ascribed to the B/PC.

In order to classify all patients, the DS(Cam) was modified by renouncement of the requirement of pancytopenia and decreased marrow cellularity in patients with mild to moderate disease. To allow for calculation of PI(Lyn), PI(Naj) and EP(Roz) the missing values of the corrected reticulocyte count in 1 patient and the

PREDICTION OF SURVIVAL 12 1

mean red cell volume (MCV) in 2 patients were sub- stituted for by the survival group medians.

Finally, all patients with haemorrhagic symptoms were assigned the value 2.5 for the variable C in the calculation of PI(Naj), as further discrimination between moderate and severe degree was not possible.

Statistical methods Qualitative variables were analysed by the log likelihood ratio statistic and quantitative variables by Mann- Whitney’s test for unpaired samples or Wilcoxon’s two-sample test (6). P = 0.05 (two-tailed) was chosen as the overall statistical significance limit. The cumulative survival was calculated according to the method of Kaplan & Meier (7).

Combinations of variables, displaying significant simultaneous association with survival, were searched for by application of stepwise linear logistic regression analysis for the case of the dichotomous dependent survival variable (8), and stepwise linear log hazard function regression analysis for the case of the quantitative dependent survival variable (9). The analyses were repeated with inclusion of variables which seemed of predictive value during preceding analyses, but which were not selected automatically. A variable was included only if the initial values were available in all patients.

RESULTS

S u r v i v a 1 The cumulative survival curve appeared biphasic with a high initial death-rate re- placed by a more moderate death-rate in pa-

tients having survived the first 4 months (Figure 1). Accordingly, survival for shorter ‘or longer than 4 months was chosen as the general dichotomous dependent survival variable.

16 patients, among whom 12 of the 23 patients with AA, died within 4 months, 1 after development of acute leukaemia, while 28 patients survived more than 4 months. A total of 34 patients (19 with AA) died, 12 after development of acute leukaemia as reported previously (5) . The causes of death in the 22 non-leukaemic deceased patients were infection (10 patients), bleeding ( 5 ) , infection and bleeding (3), severe anaemia (2), and unknown (2). 10 patients (4 with AA) were alive and had been ob- served for between 38 and 155 months. 8 of these were in complete remission and had been so for more than 2 years, while 1 had experienced .a partial and 1 no remission.

Univariate analysis The initial values and the results of the statistical analysis of variables, which were found significantly different in the short and long surviving groups, are shown in Table 1.

Anamnestic and physical findings. Only haemorrhagic manifestations at onset were sig-

Cumulative survival (%)

I I I I I 7f-

4 8 12 24 55 Months

Figure 1. The cumulative survival (k 1 SD) of 44 patients with acquired bi- or pancytopenia, calculated according to Kaplan & Meier (7).

122 CLEMMESEN ET AL

TABLE 1 Initial physical, blood and bone marrow findings found significantly different in 16 short

surviving (S 4 months) and 28.long surviving (> 4 months) patients For each group means and ranges are given of the values of quantitative variables

or numbers of patients in the separate classes of qualitative variables

Resu l t o f I Long

Var iah le (5 4 months) 1 6 patients j 28 p a t i e n t s

Physical f i n d i n g s

Haemorrhagic mani fes ta t ions ( + l o )

Blood f i n d i n g s

Thrornbocyte count ( x 10 11)

Granulocyte count ( x lo9/] )

Moriocytc count ( x 10~11)

Pancytopenia ( + l o )

Erythroblas t s ( + l o )

9

Marrow f i n d i n g s

Overall marrow c e l l u l a r i t y ( t + + / + + / + / O / - / - - l - - - )

Megakaryocytic c e l l u l a r i t y ( * + / + / O / - / - - )

Myelopoietic c e l l u l n r i t y Z ( i t / + I O / - / - - )

Eiyelopoiesis (g)'

tiyel obJ dsts (%)

Ciraiwlolmi:tic c e l l u l a r i t y 3

Granulopoiesis ( a ) 3 Hon-myeloid c e l l s (%)

( + t / + / O / - / - - )

1313

13 (2-263)

0 . 6 (0-6.3)

0.06 (0-0.5)

1412

917

2 /1 /1 /2 /2 /5 /3

0 /0 /1 /5 /10

0 / 2 / 2 / 4 / 8

3 . 3 (0.25-24)

1 (0-5)

0 I 1 / I ! 13 I8

10 (0.?5-50)

67 (5-95)

P = 0.006 (LLR) 11/17

78 (1-365) P = O.CCj.5 [!:-i')

1.2 (0-8.5) P 7 0.25 ( l L ! Y )

0.2 (0-1.8) P = 0.006 i l l - ! \ )

12/16 P = 0.00; (LLt:)

6/22 P = 0.02 (LLR!

10/1 /6 /3 /2 /2 /4

0/4/10/6/S

0/3 /16/4 /5

10.5 (0.12-27)

2 (0-8)

r l 8 l l o l o l s

34 (0-61)

27 (7-?5)

--____-- _ _ ~ . -

1. LLR = log likelihood ratio test; M-W = Mann-Whitney's test. 2. Myelopoiesis = myeloblasts, promyelocytes and myelocytes. 3. Granulopoiesis = metamyelocytes and band-shaped and polymorphonuclear

granulocytes.

nificantly more frequent in short survivors while no differences in the occurrence of infection, age, sex, and duration of onset symptoms were fbund, although the duration of onset symptoms tended to be longer in long survivors. The occurrence of a history of drug exposure was not significantly

different in short and long survivors either, irrespective of whether: (i) only agents with known marrow toxicity according to Wil- liams et a1 (10) were considered; (ii) agents with possible marrow toxicity were additionally included, or (iii) any agent were considered.

PREDICTION OF SURVIVAL 123

Blood. Short survivors significantly more frequently had pancytopenia, lower granulocyte, monocyte and thrombocyte counts, and more frequently circulating erythroblasts than long survivors. In the 15 patients initially with erythroblasts in peripheral blood, the erythroblasts made up from less than 1 % to 8 70 of the nucleated blood cells (median = 1%). The occurrence of circulating erythroblasts was not significantly associated with Hb level, reticulocyte count, signs of haemolysis or a retrospective diagnosis of AA.

The corrected reticulocyte count tended to be lower in short survivors, but this difference did not reach statistical significance. The Hb, lymphocyte count, ESR, serum gamma globulin, MCV, and the occurrence of abnormal erythrocyte and thrombocyte morphology did not differ significantly in the two survival groups.

Bone marrow. The estimated absolute cellularity of the bone marrow and its myelopoietic, granulopoietic, and megakaryocytic compartments displayed significant shifts towards lower degrees of cellularity in short survivors as compared to long survivors. Correspondingly, the relative bone marrow content of myeloblasts and myelopoietic and granulopoietic cells was significantly lower in short survivors, and the % of non-myeloid cells, mainly lymphocytes, significantly higher (for explanation, cf. foot- note of Table 1).

Only the magnitude of the erythropoietic compartment was not different in the two survival groups, neither when estimated absolute cellularity nor when relative marrow content (in %) was considered.

Multivariate analysis By repeated multiple logistic regression

analyses 2 combinations of variables displaying significant simultaneous association with short survival were found: (i) bone marrow non-myeloid cell proportion ( P =

0.004), occurrence of circulating erythroblasts ( P = 0.01), and history of any drug or chemical exposure irrespective of previously reported marrow toxicity ( P =

0.04); (ii) bone marrow non-myeloid cell proportion ( P = 0.02) and occurrence of haemorrhagic manifestations ( P = 0.03). How- ever, since the latter combination was al- ready included in PI(Lyn) and PI(Naj), only the former was further evaluated.

The discrimination between short and long survivors by the probability EP of short survival, esti.mated from the marrow non- myeloid cell % , circulating erythroblasts and history of any drug exposure, is shown in Figure 2 together with graphs displaying the variation of EP as a function of the marrow non-myeloid cell % for each of the 4 possible combinations of values of the other 2 variables.

Survival time, however, is a quantitative variable, and the reduction into a dichotomous variable implies a loss of information. Furthermore, a survival for 4 months is still a somewhat arbitrary level of discrimination between the short and long surviving group despite the biphasic survival curve. Consequently, log hazard function regression analysis of the censored survival time data was additionally applied and re- sulted in the identification of a combination of 4 variables associated significantly and simultaneously with survival: Circulating erythroblasts ( P = 0.0005), marrow non- myeloid cell proportion ( P = 0.002), any drug exposure ( P = 0.008), and age ( P =

0.01). The linear argument of the log hazard function can be regarded as a prognostic index (PI) and was calculated for each pa-

124 CLEMMESEN ET AL

100-

8 0-

60-

40-

20-

EP

0 20 40 60 80 100% Non-myeloid bone marrow cells

Figure 2. The distribution between 16 short surviving (< 4 months) (0) and 28 long surviving (> 4 months) (0) patients and graphs of the probability of survival of less than 4 months (EP), estimated from the non- myeloid bone marrow cell % (NM), occurrence of circulating erythroblasts (CE) and history of any drug exposure (DE) according to the logistic regression formula log, (EP/(l-EP)) = -2.47 + 0.055 NM + 2.57

Graph A: circulating erythroblasts (CE = l), but no

Graph B: circulating erythroblasts (CE = 1) and

Graph C: no circulating erythroblasts (CE = 0) and

Graph D: no circulating erythroblasts (CE = 0), but

C E - 2.39 DE.

history of drug exposure (DE = 0).


no history of drug exposure (DE = 0).


Cumulative survival (%)

tient from the values of these 4 variables. On arbitrary division of the patients into 3 groups of approximately equal size according to their PI values, marked differences in cumulative survival between the ' 3 groups appeared (Figure 3), demonstrating the discriminative capacity of PI.

The results of the 2 multivariate methods applied were highly consistent, all things considered. They both disclosed a significant association of increased non-myeloid cell % and occurrence of circulating erythroblasts with reduced survival, and between a history of any drug exposure and a more favourable course. In addition the log hazard function regression analysis revealed a significant contribution to reduced survival from increasing age.

Comparison of prognostic indices and criteria The capacities of the quantitative indices as well as the dichotomous criteria derived from these indices and DS(Cam) to discriminate between the patients surviving shorter and longer than 4 months were compared. Only PI was not comparable with other indices.

Figure 3. The cumulative survival, calculated according to the method of Kaplan & Meier (7), within 3 groups of approximately equal sizes, characterized by the non-myeloid bone marrow cell proportion . .

I 1 (NM; %), occurrence of circulating

history of any drug exposure I I

I I (DE; no = 0, yes = l), j 9 and age (AG; years) according to I /;c the linear argument PI of the log

LL, , I -A erythroblasts (CE; no = 0, yes = l), I ;

I I I I 4 8 12 24 'I $j Months hazard function:

PI = -0.29 + 0.0243 NM + 1.45 CE - 1.16 DE + 0.0252 AG. Group A: PI = -0.36 - 1.31 (14 patients). Group B: PI = 1.34 - 2.02 (16 patients). Group C: PI = 2.09 - 4.21 (14 patients).

PREDICTION OF SURVIVAL

P r o g n o s t i c D i a g n o s t i c D i a g n o s t i c c r i t c r i c n s p e c i f i c i t y ' s e n s i t i v i t y 2

125

D i s c r i m i n a t i v e c a p a c i t y

LT p

1. Diagnostic specificity = fraction of subjects with criterion fulfilled, who survived less than 4 months (numbers shown in brackets).

2. Diagnostic sensitivity = fraction of subjects with criterion not fulfilled, who survived longer than 4 months (numbers shown in brackets).

LT = value of log likelihood ratio test statistic ( 6 ) .

Quantitative indices. The value of the Wil- ing to this measure, E P was most dis- coxon two-sample test statistic (W2T) was criminative (W2T = 3.88; P = O.OoOl), used as a measure of the discrimination followed by PI(Naj) (W2T = 3.32; P = between short and long survivors. Accord- 0.001) and PI(LYn) (W2T = 2.98; P =

126 CLEMMESEN ET AL

0.003), while EP(Roz) did not discriminate significantly between short and long survivors (W2T = 1.46; P = 0.14). Even when only the 13 subjects fulfilling the original criteria of entrance of Rozman et a1 (1981) were included and survival for shorter or longer than 3 months was tested, EP(Roz) was not significantly discriminative. Apart from (EP(Roz), any of these quantitative indices were more discriminative than the most discriminative single variable, the marrow non-myeloid cell %, the W2T of which was 2.76 ( P = 0.006).

Dichotomous criteria. E P found in this study was arbitrarily reduced into the dichotomous criterion E P >/< 0.5 and compared with the criteria proposed by Camitta et a1 (l), Lynch et a1 (2), Najean & Pecking (3) and Rozman et a1 (4) (Table 2). The comparisons comprised as well all patients as subgroups of patients fulfilling the original criteria of selection of Najean & Pec- king (3), i.e. B/PC including aregenerative anaemia initially, Lynch et a1 (2), i.e. AA retrospectively, and Camitta et a1 (1) and Rozman et a1 (4), i.e. AA initially. The discriminative capacity of these criteria com- prises both the diagnostic specificity, i.e. the probability of a correct prediction of survival for less than 4 months, and the diagnostic sensitivity, i.e. the probability of a correct prediction of survival for longer than 4 months, which are given in Table 2. As a simultaneous measure of both these diagnostic properties the value of the log likelihood ratio test statistic (LT) was calculated (Table 2). It appears, that the specificities of short survival of EP > 0.5, PI(Lyn) > 0.041/ 0.033 and severe DS(Cam) did not differ substantially, while the specificities of PI(Naj) < 25 and EP(Roz) > 0.5 were below this level. However, when applied

only to patients fulfilling the original criteria of entrance, the specificities of these 2 criteria improved to the level of the other criteria. The most sensitive criterion was PI(Naj) < 50 followed by PI(Lyn) > O.OOO/ 0.033 and EP > 0.5, while DS(Cam) and EP(Roz) > 0.5 were less sensitive and displayed severe deterioration of sensitivity upon omission of patients not fulfilling the original criteria of entrance. The specificity and sensitivity of the criteria derived from PI(Lyn) were substantially unchanged in patients with AA retrospectively, and also specificities and sensitivities of E P > 0.5 appeared practically invariant on restriction of application to only patients with AA retrospectively or initially. The generally less reliable estimates of specificity and sensitivity in smaller subgroups of the patient material, especially the small subgroups of patients with A A initially, should, however, be kept in mind.

DISCUSSION

The biphasic survival curve in this material of 44 patients confirms the existence of a subgroup of patients with survival < 4 months, most of whom (12/16) fulfil the retrospective criteria of AA. Furthermore, this study supports this subgroup being al- ready identifiable with a considerable degree of precision at the first admission and this prediction being profitably based on multiple variables.

A biphasic survival curve in AA with a clear deflection after 3 to 4 months has been pointed out previously (2, 11). A de- crease of the death rate within the first 2 months up to 1 year after diagnosis has been recognized earlier (12-15).


The prognosis of AA has been the sub- ject of a considerable number of often not comparable reports. In the following, we will mainly refer to reports in which the dependent survival variable was dichotomous as regards levels of survival for shorter and longer duration than from a few to 6 months.

The association between initial haemorrhagic manifestations, granulocyte and thrombocyte count and marrow myeloidl non-myeloid cell proportion, and short survival as found in this study is in agreement with similar findings in previous studies (2- 4, 12, 15-18). Furthermore, the lack of predictive value of the initial Hb is consistent with previous reports (4, 16, 17).

The results of previous studies of the prognostic significance of age and sex are more contradictory. Lewis (12) and Najean & Pecking (11) found an impairment of survival with increasing age, while others, (2, 4) did not find any correlation. In this study age alone did not affect survival, but combined with the effects of marrow non- myeloid cell proportion, circulating erythroblasts and a history of drug exposure, increasing age was associated with shorter survival. Our finding of no prognostic significance of sex is concordant with one previous study (4), while one showed a more favourable outcome in females (2), and another in males (3).

Most studies have failed to disclose any difference in survival between drug-induced and idiopathic AA ( 2 4 , 12). Only Vincent & de Gruchy (13) found a lower mortality rate in drug-induced than in idiopathic AA, supporting the prognostic significance of a history of any drug exposure in this study, while Camitta et a1 (19) in contrast found a longer survival in idiopathic AA than AA associated with suspected etiology.

The prognostic significance of occurrence of circulating erythroblasts found in this study has apparently not been studied previously. Furthermore, the value of semi- quantitative estimates of the absolute cellularity of the bone marrow and its compartments has only been studied infrequently and most often with disappointing results (11, 16). Only te Velde & Haak (20) have found semi-quantitative cellularity scores useful in prediction of short survival in AA.

This study could not demonstrate the significant predictive value of the duration of onset symptoms and the initial reticulocyte count, which have been consistent findings in some studies (2-4, 16, 17). Additionally a few studies, applying diverging dependent survival variables, have supported the predictive value of the reticulocyte count (21-23). However, this discrepancy is not serious, since trends towards lower values in short survivors were found for both of these 2 variables. The prognostic significance of the MCV found in 2 previous studies (4, 19) could not be verified in this study either.

In this study, the application of multivariate analyses lead to the identification of prognostic variables containing significant independent predictive information, even among variables not associated with survival according to univariate analyses. Fur- thermore, the resulting multivariate prognostic indices and criteria gained in predictive capacity compared with the single variables, by which they were made up. The finding of a simultaneous association of 2 variables, viz. bone marrow non-myeloid cell proportion and haemorrhagic manifestations, with short survival, confirms the validity of PI(Lyn) and PI(Naj), in which these variables enter together with other variables. Accordingly this combination of

128 CLEMMESEN ET AL

variables was not considered of any further independent interest. However, the value of the indices E P and PI identified in this study, derives from the following variables drug exposure, age, and circulating erythroblasts. These have not previously been found of prognostic significance in relation to multivariate indices.

Considering both the quantitative indices and the derived and original dichotomous criteria of short survival it may be con- cluded that PI(Naj), PI(Lyn), our EP, and DS(Cam) in the present patient material were all significantly predictive and approximately equally discriminative. Further- more, PI(Naj) and PI(Lyn) appeared rather robust to application to patients fulfilling more extensive criteria of selection, while our EP on the other hand seemed to retain its discriminative capacity in the more re- stricted groups of patients with AA retrospectively or initially. Thus, our study supports the predictive value of PI(Lyn) and DS(Cam) which has been questioned by te Velde & Haak (20) and van Hengstum et a1 (24), respectively. Conversely, EP(Roz) was not found significantly predictive, but PI(Lyn) also appeared less discriminative than originally reported (2). In neither case could the deficient predictive capacities be sufficiently explained by different criteria of patient selection. This emphasizes that prognostic indices and criteria as pointed out by Rozman et al (4), tend to be less precise in populations other than those, from which they were originally derived. Accordingly, the use of E P and PI in clinical practice in predicting survival needs to be ascertained on other patient materials.

Although some of the existing prognostic indices and criteria may be considered useful for clinical practice and trials, further improvement of the predictive precision is

desirable. However, whether this is at all possible or whether the residual lack of precision of the existing indices and criteria is entirely due to random error variation is for the moment unknown. The search for an answer to this question would probably re- quire: (i) investigation of large materials of patients accumulated multicentrically as in the study of Najean & Pecking (3, 11); (ii) generally accepted and prospectively applicable entrance criteria; ( 5 ) inclusion in the study of variables not directly repre- senting the degree of severity of B/PC and marrow hypoplasia, for instance the occurrence of circulating erythroblasts and a history of any drug exposure found valuable in this study; and finally (iv) application of an adequate multivariate statistical analysis, the log hazard function regression analysis of Cox (8) applied in this study and previously by Camitta et a1 (19) being probably most efficient due to the independence of data distribution and the ability to deal with the original quantitative censored survival data.

1

REFERENCES

Camitta BM, Rappeport JM, Parkman R, Nathan DG. Selection of patients for bone marrow transplantation in severe aplastic anemia. Blood 1975; 45 :355-63.

2. Lynch RE, Williams DM, Reading JC, Cartwright GE. The prognosis in aplastic anemia. Blood 1975 ;45:5 17-28.

3. Najean Y, Pecking A. Prognostic factors in acquired aplastic anemia. A study of 352 cases. Am J Med 1979;67:56471.

4. Rozman C, Marin P, Granena A et al. Prognosis in acquired aplastic anaemia. A multivariate statistical analysis of 80 cases. Scand J Haematol 1981 ;26:321-9.

5. Clemmesen L, M0rk Hansen M, Mortensen L, Thiede T, Videbrek Aa. The applicability of sim- ple blood and bone marrow variables in prediction


of acute leukaemia in patients with acquired bi- or pancytopenia. Scand J Haematol 1981;27:193-200.

6. Sokal RR, Rohlf FJ. Biometry. San Francisco: WH Freeman and Company, 1969.

7. Kaplan EL, Meier P. Nonparametric estimation from incomplete observations. J Am Statist Assoc 1958;53:457-81.

8. Cox DR. The analysis of binary data. London: Methuen, 1970;9M.

9. Cox DR. Regression models and life-tables. J Roc Statist SOC B 1972;34:187-202.

10. Williams DM, Lynch RE, Cartwright GE. Drug- induced aplastic anemia. Semin Hematol 1973;lO: 195-223.

11. Najean Y, Pecking A. Le prognostic a court terme des aplasies mkdullaires de l’adulte. Nouv Presse Med 1977;6:3083-8.

12. Lewis SM. Course and prognosis in aplastic anaemia. Br Med J 1965;i:1027-31.

13. Vincent PC, De Gruchy GC. Complications and treatment of acquired aplastic anaemia. Br J Haematol 1967;13:977-99.

14. Bottiger LE, Westerholm B. Aplastic anaemia. I. Incidence and etiology. Acta Med Scand 1972;192:

15. Davis S, Rubin AD. Treatment and prognosis in aplastic anaemia. Lancet 1972;i:871-3.

16. Najean Y, Bernard J, Wainberger M, Dresch C, Boiron M, Seligmann M. Evolution et prognostic des pancytopknies idiopathiques. Etude de 116 observations. Nouv Rev Fr Hematol 1965;5:639-56.

315-8.

17. Hartwich G, Schwabel H-J, Sailer D, Lutz H. Untersuchungsergebnisse bei 50 Patienten mit Pan- myelopathie. Med Klin 1973;68:765-71.

18. Frisch B, Lewis SM. The bone marrow in aplastic anaemia: Diagnostic and prognostic features. J Clin Pathol 1974;27:23141.

19. Camitta BM, Thomas ED, Nathan DG et al. A prospective study of androgens and bone marrow transplantation for treatment of severe aplastic anemia. Blood 1979;53:50&14.

20. te Velde J, Haak HL. Aplastic anaemia. Histolog- ical investigation of methacrylate embedded bone marrow biopsy specimens; correlation with survival after conventional treatment in 15 adult patients. Br J Haematol 1977;35:61-9.

21. Heimpel H, Rehbock C, van Eimeren W. Verlauf und Prognose der Panmyelopathie und der iso- lierten aplastischen Anamie. Eine retrospektive Studie an 70 Patienten. Blut 1975;30:235-54.

22. Lohrmann H-P, Kern P, Niethammer D, Heimpel H. Identification of high-risk patients with aplastic anaemia in selection for allogeneic bone-marrow transplantation. Lancet 1976;ii:647-50.

23. Hellriegel K-P, Zuger M, Gross R. Prognosis in acquired aplastic anemia. An approach in the selection of patients for allogeneic bone marrow transplantation. Blut 1977;34:11-8.

24. van Hengstum M, Steenbergen J, Haanen C. Clin- ical course in 28 unselected patients with aplastic anaemia treated with anabolic steroids. Br J Haematol 1979;41:32>33.

Scand J Haematol(l984) 32 9

Documents

Prediction of Survival in Adults with Acquired Bi- or Pancytopenia