British Journal ofHaernatology. 1985, 61, 357-370
Classification of anaemia on the basis of ferrokinetic parameters
G . BAROSI, M. CAZZOLA, C. BERZUINI,* S. QUAGLINI' A N D M. STEFANELLI* Dipartimento di Medicina Interna e Terapia Medica, Policlinico S. Matteo, and Dipartimento di lnformatica e Sistemistica, Strada Nuova 106/C, 271 00 Pavia, ltaly
Received 3 August 1 9 8 4 ; accepted for publication 9 January 1 9 8 5
SUMMARY. Quantitative information on abnormalities of erythropoiesis and mechanisms of anaemia has been obtained in 1 3 6 anaemic patients by means of ferrokinetic studies. To derive a functional classification of anaemia based on ferrokinetic parameters, agglomerative hierarchical cluster analysis and princi- pal coordinate analysis were utilized as techniques for unsupervised classifica- tion. Two main clusters were found and named anaemia with low potential erythropoiesis and with high potential erythropoiesis, since the most discri- minant parameter between them was total erythroid iron turnover, a measure of total erythropoietic activity. A value of total erythropoiesis equal to 4 times the normal was found to discriminate these two types of anaemia in 94% of cases. Within the group with low potential erythropoiesis, three clusters showing different qualitative disturbances of erythropoiesis were singled out. Among patients with high potential erythropoiesis, two clusters were found. A value of effective erythropoiesis equal to 2 . 5 times the normal was shown to have a high discriminant power between these clusters. This threshold level distinguished between patients having ineffective erythropoiesis or peripheral haemolysis as the major mechanism of anaemia. The present functional classification of anaemia provides a complete picture of the different pathogene- tic mechanisms and may represent the basis for a more rational diagnostic approach to erythroid disorders.
Anaemia may be functionally classified as hypoproliferative or hyperproliferative on the basis of the bone marrow picture (Finch, 1982; Dancey et al, 1976). Reticulocyte count or index (Hillman & Finch, 1969), in addition, provides a simple estimation of the efficiency of red cell production. To derive a more accurate quantitation of erythroid proliferation and its efficiency a variety of kinetic studies such as bilirubin turnover (Engstedt et al, 1967; Berk et
Correspondence: Dr G. Barosi, Dipartimento di Medicina Interna e Terapia Medica, Policlinico S. Matteo, 27100 Pavia, Italy.
3 5 8 G. Barosi et al
al, 1976; Samson et al, 1976), carbon monoxide production (Coburn et al, 1968; Lundh et al, 1975) and fecal urobilin excretion (Crosby, 1950). have been employed.
A valuable improvement in our understanding of the pathogenesis of anaemia has been provided by ferrokinetic studies (Finch et al, 1970; Cavil1 & Ricketts, 1980; Stefanelli et ul, 1982), which allow the quantitation of total, effective and ineffective erythropoiesis and mean red cell lifespan. These studies have shown that, first, the rate of erythroid proliferation can range from a negligible level to as much as 15 times the normal (Pootrakul et al, 1981) according to the severity of the anaemia and the proliferative capacity of the erythroid marrow. Second, a mild degree of ineffective erythropoiesis is present in healthy bone marrow and it may reach 90% of total erythropoiesis in some diseases (Stefanelli et al, 1982; Barosi et al, 1981; Cazzola et al, 1982). Third, the major mechanism producing anaemia may be different across patients with the same disorder, and ferrokinetic studies allow it to be identified in individual patients (Barosi et al, 1979).
In the present work, we have analysed ferrokinetic results obtained in 136 anaemic patients with the aim of defining a functional classification of anaemia based on the quantitative evaluation of erythropoiesis. Then, we have tested the feasibility of using simple haematological parameters for discriminating groups of patients according to the classifica- tion obtained.
MATERIALS A N D METHODS
In the last 10 years more than 300 patients referred to our Clinical Department underwent an erythrokinetic study, i.e. ferrokinetics combined with red cell survival study. Due to the aims of the present work, only 136 cases were utilized. The selection criteria were the following: (a) anaemia was present at the time of study, that is haemoglobin concentration was lower than 14 g/dl in males and 12 g/dl in females; (b) a definite clinical diagnosis had been established; (c) patients were not strictly transfusion dependent. In no case had the patient been transfused in the 2 months before the study: (d) no treatment capable of modifying erythropoiesis had been given before the study; (e) none of the patients had a multisystem disorder likely to influence the degree of anaemia.
Haemoglobin concentration (Hb), mean corpuscular haemoglobin concentration (MCHC) and mean cell volume (MVC) were measured by Coulter S Counter. Reticulocyte count was measured by standard technique (Dacie & Lewis, 1975). Reticulocyte index was then computed according to Hillman & Finch (1969). Serum iron (SI) and transferrin saturation (TS) were determined by colorimetric analyses (ICSH Panel, 1 978a, b). Bilirubin concentra- tion was obtained by automated technique. In each case bone marrow aspirate and/or trephine biopsy were used to establish diagnosis. Summary statistics of haematological data are reported in Table I, where clinical classes were defined according to Bothwell et a1 (1 9 79).
Classijfication of Anaemia 359
Table I. Summary statistics for haematological parameters within clinical classes
Haematological parameters No. of Age Hb MCV MCHC Ketics Serum iron Transf. sat. Bilir.
Disease cases (yr) (g/dl) (fl) (g/dl) (10'/1) (pmol/l) (%) (ms/l) -
48.0 7.9 101.0 31.8 37.5 38.8 73.0 26-71 5'2-11'4 88-121 30.2-34.3 2-326, 25.0-62'7 53-95 59.1 8.1 101.4 31.1 67.0 25.1 50.0
40-77 4.9-12.2 82-130 30.0-32.9 10-300 1 0 * 0 4 4 * 0 21-81
66.5 6.7 89.5 31.8 36.5 25.5 45.5 61-72 6.5-6.9 89-90 28.8-34.8 23-50 11.0-40.0 21-70
59.2 8.9 99.4 32.4 50.6 24.1 47.9 14-77 6.2-12.3 86-123 28.8-34.8 16-96 5.4-54.1 13-86
54.0 10.5 94.6 32.4 62.0 19.6 ma.
34.5 8.1 71.5 28.8 67.0 4.1 5.6 7-63 5'7-10'9 59-90 27.1-32'6 7-166 2.0-8'0 3-11 19.0 10.0 96.0 32.9 39.6 30.0 57.0 8-40 7.2-12.0 85-1 12 29'3-34'3 8-76 10'7-38.9 22'8-75 66.0 7.4 104.0 31.2 36.5 27.8 56.0
47-69 8.2-12.8 67-105 29.6-35.0 9-341 10'7-35'4
50-80 6'0-8'1 96-117 29'5-34'0 7-127 1 0 ' 4 4 4 . 7 19-80 34.5 8.3 71.1 28.5 237.8 32.3 65.7 6-54 5'1-12'0 57-78 24.8-31.9 46-514 23'2-48.6 49-85 36.5 12.0 91.0 35.4 237.4 20.8 38.3
1 4 4 9 10'6-13.0 83-106 25'5-36'9 81-368 10'3-32'2 30-51 46.7 8.5 107.5 32.2 249.7 14.3 n.a.
31-58 6'4-9'7 93-130 29.0-35.5 181-300 10.3-21.6 55.0 10.0 88.8 32.1 114.0 19.9 28.0
28-77 5'2-13'6 74-102 29'1-34'8 2 0 4 5 0 6 . 7 4 4 ' 7 1 5 4 6
9.9 6.3-1 7.0
13.8 4'9-3 8 9
7.0 5 4-9 .0
8.7 3.1-1 7.8
8.3 2.7-1 3.1
. 5.8 3.8-7.8
30.5 22.1-3 8.7
13.6 8.8-1 8.4
8.8 4.3-2 2 ' 0
n.a., not available.
The experimental protocol and methods for the analysis of erythrokinetic data have been extensively presented elsewhere (Stefanelli et ul, 19 82) . Subject's transferrin, specifically labelled with 59Fe (ferric citrate), was injected intravenously and blood samples were taken according to a suitable sampling schedule for 14 d. The plasma volume was measured by dilution of 59Fe transferrin and red cell volume (RCV) was measured by "Cr method of ICSH (1 9 71). The plasma 59Fe disappearance and red cell utilization curves were used to estimate the parameters of a multicompartment model of iron kinetics. From these estimates, the following ferrokinetic parameters were calculated: total erythroid iron turnover (TEIT), a measure of total erythropoiesis; ineffective iron turnover (IIT), a measure of ineffective
3 60 G. Barosi et a1
erythropoiesis. It may be more usefully expressed as a percentage of TEIT and indicated as IW%; red cell iron turnover (KCTT), a measure of effective erythropoiesis: mean red cell life span (MKCI,), the inverse of which provides an estimate of the rate of peripheral haemolysis.
In this paper, classification is considered to be the process of allocating patients of the studied sample to initially undefined classes, called clusters, so that patients in the same cluster are, from the point of view of the functional behaviour of erythropoiesis, similar to one another. Five logical steps can be distinguished in this process:
1. Selection of parameters for the classiJication. 2 . Preliminary tests of rridtiriorrrialitg or1 the global data set. They indicate presence of
3. Deterniination of clusters by partitioning of the patients sample. 4. Characterization of clusters obtained. 5. Cluster rhoire arid validation.
Steps 1-5 are described in the following. Step 1. After gaining experience via several cluster analyses based on different subsets of
ferrokinetic and haematological variables, the final classification was based only 011 estimated ferrokinetic parameters TETT, RCIT, IITX and MRCL.
Step 2. Simple tests of multinormality and data exploration techniques (Gnanadesikan 1977) were at first carried out on the data. It was found that data contained strong non-normal features, allowing the rejection of the hypothesis of multinormality.
Step 3. Agglomerative hierarchical cluster analysis (Everlitt, 19 74) and principal components analysis (Gower, 1966) were utilized to determine the clusters.
The agglomerative hierarchical clustering algorithm can be described using a diagram- matic representation, as that shown in Fig 1, called a dendrogram. Initially, each patient is considered a separate cluster. The algorithm starts by grouping the two nearest patients. This process is repeated by fusing iteratively the two closest clusters, thus reducing at each step the number of clusters by one, until all patients are clustered together
Since hierarchical cluster analysis is likely to yield artefacts when clusters are not well isolated or unusually shaped, principal components analysis (PCA) was used for checking by eye inspection clusters found and for local corrections of the classification.
Step 4. Linear discriminant anaiysis (Jennrich, 19 77) on the basis of the same parameters used for the classification has been applied to clusters obtained, as an informal indicator of which parameters have contributed most to cluster formation. Hence a characterization of clusters has been obtained.
All results of analyses included in step 4 provided a tool for the interpretation of dusters from the patho-physiological point of view.
Step 5 . The number of clusters to be retained, i.e. the cut level of the dendrogram, was chosen. This was made judging the clusters on the basis of results of step 4 and external contextual knowledge, to see whether they corresponded to a rational set of different functional patterns of erythropoiesis.
potentially interesting pattern in the data, and motivate subsequent steps of the analysis.
Classification of Anaemia 361
Based on ferrokinetic parameters, the classification represented by the dendrogram shown in Fig 1 was obtained.
a, u c m
50 - sl L
c a, G
I - A1
13/13 A A 17/17 RAE6 2/2 CMML 10/10 IDA 2/11 HCL 7/35 MMM
a / a R A
8/a HS 218 CDA
4/4 AIHA 1 11/35 MMM 3/11 PASA 4 /9 THAL 6/8 CDA 5 /9 THAL 11/35 MMM 8/11 PASA Fig 1. Dendrogram for the classification of anaemias based on ferrokinetic parameters. A denotes anaemias with low-potential erythropoiesis and B anaemias with high-potential erythropoiesis. Boxes represent clusters a t lowest level considered meaningful. Inside boxes, cluster compositions are reported in terms of clinical diagnoses (n/m means n out of m patients). Labels of clinical diagnoses are explained in Table I.
Separation into anuemiu with low potential and high potentiul erythropoiesis
The two biggest and farthest apart clusters joining at the root of the dendrogram, denoted as A and B, contained 74 and 62 patients, respectively. Table I1 shows summary statistics for haematological and ferrokinetic parameters in these clusters. TEIT was found to be the most discriminant parameter between the two clusters: it ranged from 59 to 3 78 pmol/l blood/d in A and from 292 to 1474 pmol/I blood/d in B. A threshold value of 350 pmol/l blood/d for TEIT allowed 126 out of 136 patients (94%) to be correctly allocated.
Discriminant analysis was applied between cluster A and B using only haematological parameters. The most discriminant parameter was found to be bilirubin concentration, which had a mean value of 8.8 f 5.3 mg/l in cluster A and 23.4 =t 15.9 mg/l in cluster B and allowed 77% of cases to be correctly allocated. The remaining haematological parameters were unable to improve significantly the separation.
F a 16
!2 g E
E 3 B
3 64 G. Sarosi f t a1
Clustering of anaemias with low potential erythropoiesis
The cluster of anaemias with low potential erythropoiesis was further subdivided into three well-separated clusters, denoted by A l , A2 and A3 (see Fig 1). Summary statistics for haematological and ferrokinetic parameters within these clusters are presented in Table 111.
The strongest separation appeared to be that between the group of nine cases falling in A 1 and the remaining 65 cases falling in A2 or A3. MRCL was found to be the most discriminant parameter: its mean value was 1 0 5 f 13.9 d in A1 and 51.2 f 13 .6 d in A2 and A3 pooled together. Cluster A1 contained the majority of cases of hairy cell leukaemia (HCL) (nine out of 11) and no other patient. Pattern of erythropoiesis in these patients is very close to that of normal subjects.
The separation betwen A2 and A3 was mainly explained by TEIT, which ranged from 1 to 2.5 times the normal in A2 and from 2.2 to 3.5 times the normal in A3. Cluster A2 included all patients with aplastic anaemia (AA), refractory anaemia with excess of blasts (RAEB), chronic myelomonocytic leukaemia (CMML) and IDA, besides seven patients with myelo-
N L L
2- 0 .-
.Cluster A 1
0 Cluster A 2
0 Cluster A 3
0 0 0
00 0 00
0 0 0 0 000 a
0 0 0 00000 0
0 ooo800 00 O 0 o o o
0 000 0
m * .. m .
First Discriminant Function F1 Fig 2. Diagram of patients with low-potential erythropoiesis on the plane defined by two linear discriminant functions for the clusters A l , A2 and A3. This is the plane where maximum distances between centroids (represented by asterisks) with respect to within-clusters dispersion are attained. The two discriminant functions are defined by the following linear combinations of ferrokinetic parameters: E'1=0'001 TEIT-0.00986 KCIT-0.2145 IIT0/,+0.0309 MRCL-0.259 F2 = -0.0183 TEIT+0*00303 KCIT-tO.03496 IIT0/,-0*03575 MKCL+ 3.506 Linear discriminant functions computed using only MKCL and TEIT allowed an almost equal degree of separation among the three clusters.
Classification of Anaemia 365
fibrosis with myeloid metaplasia (MMM), and two with HCL. Cluster A3 included all patients with refractory anaemia (RA) and eight patients with MMM.
The haemoglobin concentration was not significantly different in the three clusters (F= 2.36, P> 0.05). Using ferrokinetic parameters were found two discriminant functions able to allocate correctly all cases but two as shown in Fig 2. On the contrary, haematological parameters revealed a poor discriminant power among the three clusters.
Clustering of anaemias with high potential urgthropoiesis
As shown by the dendrogram reported in Fig 1 , cluster B was subdivided into two finer clusters, denoted by B 1 and B2. Summary statistics for haematological and ferrokinetic parameters are reported in Table IV. KCIT was found to be the best parameter for discriminating B1 (RCIT= 304 f 95 pmol/l blood/d) from B2 (RCIT= 128 f 3 7 pmol/l blood/d). In Fig 3 the scatter diagram of the patients belonging to B 1 or B2 is presented in the plane of the two variables RCIT and (100 - IITY,), a measure of efficiency of erythropoiesis. Therefore, such two clusters can be well defined as anaemias with effective erythropoiesis ( B l ) and with ineffective erythropoiesis (B2). A linear discriminant function in such a plane was able to separate the two groups with only one misallocated patient. Misallocations were not less than 10% using IIT% and MRCI, separately, but it reduced to 6% when the two parameters were used jointly.
Table IV. Ferrokinetic classification of anaemia
A. Anaemia with low potential erythropoiesis A l . True erythroidfailure
A2. Erythroid failure with moderate wiaturative disturbances Hairy cell leukaemia
Iron deficiency anaemia Refractory anaemia with excess of blasts Chronic myelomonocytic leukaemia Aplastic anaemia
Refractory anaemia Idiopathic myelofibrosis
A3. Erythroid failure with severe maturative disturbances
B. Anaemia with high potential erythropoiesis Bl . With inefective ergthropoiesis
Sideroblastic anaemia Congenital dyserythropoietic anaemia Idiopathic myelofibrosis Thalassaemia syndromes
Hereditary spherocytosis Autoimmune haemolytic anaemia Thalassaemia syndromes Congenital dyserythropoietic anaemia
B2. With effective erythropoiesis
A Y I- - I
8 50- v 4 0 - - v)
u) Q) 30- 0 P
0, 5 20- 1. Q)
0 > 0 c 10- 0) 0
L - .- .- r c
G. Barosi et a1
3 0," O
0 O 0
4 t e e
0 0 0 0
me. m e
8 0 0
0 0 0
0 Cluster BI Cluster a2
5 4 50 100 200 300 500 700
Red cel l i r o n turnover (prnol / l blood/day)
Fig 3. Scatter diagram of efficiency of erythropoiesis versus red cell iron turnover for patients with high potential erythropoiesis. Open and full circles denote membership of clusters B1 and B2 respectively (see Fig 1) .
Cluster B1 included all patients with hereditary spherocytosis (HS) and the patients with autoimmune haemolytic anaemia (AIHA) and four out of nine patients with thalassaemia (THAL), three out of 11 with primary acquired sideroblastic anaemia (PASA), two out of eight with congenital dyserythropoietic anaemia (CDA) and 11 out of 35 with MMM. All these last patients presented the lowest values of MKCL within their own clinical group. Cluster B2 included the remaining eight out of 11 patients with PASA, six out of eight with CDA, five out of nine with THAL and 11 out of 3 5 with MMM. Either reticulocyte count or index allowed to separate B1 from B2 with more than 30% of misallocated cases.
This work presents the results obtained by applying classification techniques to ferrokinetic parameters estimated in anaemic patients. In such patients a ferrokinetic study had been carried out for clinical purposes. Thus, our sample neither included all different types of anaemia nor reproduced their relative frequency in the population. Nevertheless, as the data analysis is purely descriptive, this does not jeopardize the main purpose of the present work, that is to single out basic patterns of erythropoiesis in anaemia.
The results of the classification process was the subdivision of the patients into five clusters, whose description in terms of clinical diagnosis is given in Table IV. Two main clusters were found and named anaemia with low potential and high potential erythro- poiesis, since the most discriminant parameter between them was total erythroid iron
Classification of Anaemia 367
L 6 8 10 12 1 b
Haemoglobin concentrat ion ( 9 /dl ) Fig 4. Decision boundary for establishing whether a patient presents an anaemia with low- or high-potential erythropoiesis.
turnover, a measure of total erythropoietic activity. The primarily impaired erythroid proliferation, combined with a moderate defect of maturation, determines the occurrence of anaemia in patients with low potential erythropoiesis. In such cases erythropoiesis is inadequately increased to compensate for the ineffective erythropoiesis and peripheral red cell destruction. On the other hand, erythropoiesis is highly stimulated by the anaemia in patients with high potential erythropoiesis, but most of this effort is wasted by intramedullary and/or extramedullary red cell destruction.
A value for total erythropoiesis equal to 4 times normal was found to discriminate patients with low potential erythropoiesis from those with high potential erythropoiesis in 94% of cases. Thus, when erythropoiesis is lower than 4 times normal, anaemia can be defined as due to inadequate erythropoiesis. A more precise rule to establish whether a patient presents an anaemia with a low or high potential erythropoiesis can be derived by defining a decision boundary in the plane of total erythropoiesis standardized to the normal value and haemoglobin concentration. Such a decision boundary is shown in Fig 4 and was obtained by calculating the curves of the lower fifth percentile of total erythropoiesis in patients with high potential erythropoiesis and the upper ninety-fifth percentile in patients with low potential erythropoiesis as a function of haemoglobin concentration.
Patients with anaemia due to defective stem cell proliferation, i.e. aplastic anaemia, bone marrow infiltration by lymphoid ceils, i.e. hairy cell Ieukaemia, and iron deficiency fell in the low potential erythropoiesis cluster. On the other hand, patients with disorders affecting maturing erythroblasts, i.e. congenital dyserythropoietic anaemia type I1 and thalassaemia, or circulating red cells, i.e. hereditary spherocytosis and autoimmune haemolytic anaemia, fell in the high potential erythropoiesis cluster.
Myelodysplastic syndromes and idiopathic myelofibrosis appeared to be heterogeneous diseases. About one third of cases with idiopathic myelofibrosis were classified as having low potential erythropoiesis and the remaining as high potential erythropoiesis. In fact,
368 G. Burosi et al
idiopathic myelotibrosis is known to be a myeloproliferative disease with either hypo- or hyperproliferative marrow. In a previous study (Barosi et a!, 1981) we analysed by classification techniques a subsample of patients with idiopathic myelofibrosis. Three clusters of patients were singled out, which were found to be a valuable tool for providing a prognostic indication of the disease. Such clusters are in agreement with classification results obtained in the present work.
Refractory anaemia, with or without excess of blasts, and chronic myelomonocytic leukaemia, were classified as anaemia with low potential erythropoiesis. Patients with sideroblastic anaemia, on the contrary, fell in the cluster of anaemia with high potential erythropoiesis. Myelodysplastic syndromes give an example of the validity of these two patterns, defined on the bases of ferrokinetic parameters, for a correct classification of disorders with a hypercellular marrow (Cazzola et 01, 1982).
The three clusters within anaemia with low potential erythropoiesis may be viewed as different behaviours of erythropoiesis in the presence of erythroid failure. The common feature of erythroid failure seems to be that any erythropoietic effort sets into motion maturation disturbances which counteract red cell production. Only at normal or nearly normal levels of erythroid proliferation, erythropoiesis proceeds ordinately.
Within the group of patients with high potential erythropoiesis, two clusters were found. Discriminant analysis indicated that effective erythropoiesis was the most discriminant parameter between the two clusters, since all cases but one could be correctly classified using a threshold level of 2 . 5 times the normal (see Fig 3 ) . The mean values for mean red cell lifespan and per cent ineffective iron turnover were significantly different between the two clusters. Nevertheless, each of these parameters, taken one by one, did not separate these groups with less than 10% of misallocations. As a matter of fact, in typical maturation disorders we found values of mean red cell lifespan lower than 40 d and in haemolytic disorders values of ineffective erythropoiesis greater than 50%.
Some cases of congenital dyserythropoietic anaemia, thalassaemia and primary acquired sideroblastic anaemia, conditions clinicaIly defined as dyserythropoietic, were classified as anaemias mostly due to peripheral haemolysis. Therefore, it is apparent that the same mechanism may be responsible for both intramedullary erythroblast destruction and peripheral red cell breakdown. Thus the different erythrokinetic pattern could depend on the stage of erythroid development at which the disturbance occurs.
The well-established clinical classification of anaemia is based on morphological, aetiological or pathogenetic criteria. A functional classification might be a better basis for a general understanding of the mechanisms producing anaemia. In the present study we showed that such classification can be derived from ferrokinetic parameters.
ABBREVIATIONS IJSED IN THIS PAPER
Hb, haemoglobin concentration: MCHC, mean corpuscular haemoglobin concentration; MCV, mean cell volume; SI, serum iron; TS, transferrin saturation: AA, aplastic anaemia: MDS, myelodysplastic syndromes; KA, refractory anaemia; KAEB. refractory anaemia with excess of blasts: CMML, chronic myelomonocytic leukaemia; HCL, hairy cell leukaemia: IDA,
Classijcation of Anaemia 369
iron deficiency anaemia; CDA, congential dyserythropoietic anaemia type 11; PASA, primary acquired sideroblastic anaemia; THAL, fl-thalassaemia: HS, hereditary spherocytosis; AIHA, autoimmune haemolytic anaemia; MMM, myelofibrosis with myeloid metaplasia; TEIT, total erythroid iron turnover; IIT, ineffective iron turnover; IIT%, per cent ineffective iron turnover; RCIT, red cell iron turnover; MRCL, mean red cell life span: RCV, red cell volume; PCA, prinicipal coordinate analysis.
Major support for this research was provided by a grant from the Minister0 della Pubblica Istruzione.
BAROSI. G., CAZZOLA, M., FRASSONI, F., ORLANDI, E. & STEFANELLI, M. (1 981 ) Erythropoiesis in myelofibrosis with myeloid metaplasia: recog- nition of different classes of patients by eryth- rokinetics. British Journal of Haematology, 48,
BAROSI, G., CAZZOLA, M., STEFANELLI, M. & ASCARI, E. (1 9 79) Studies of ineffective erythropoiesis and peripheral haemolysis in congential dyser- ythropoietic anaemia type 11. British Journal of Haematology, 43, 243-250.
BERK. P.D.. BLASCHKE, T.F., SCHARSCHMIDT, B.F., WAGGONER, J.G. & BERLIN, N.I. (1976) A new approach to quantitation of the various sources of bilirubin in man. Journal of Laboratory and Clinical Medicine, 87, 767-780.
BOTHWELL, T.H., CHARLTON. R.W., COOK, J.D. & FINCH, C.A. (1979) Iron Metabolism in Man. Blackwell Scientific Publications, Oxford.
CAVILL, I. & RICKETTS, C. (1980) Human iron kinetics. Iron in Biochemistry and Medicine, 11 (ed. by A. Jacobs and M. Worwood). pp. 5 73-604. Academic Press, London.
CAZZOLA. M., BAROSI, G.. BERZUINI, C., DACC~, M., ORLANDI. E., STEFANELLI, M. & ASCARI, E. (1982) Quantitative evaluation of erythropoie- tic activity in dysmyelopoietic syndromes. Bri- tish Journal of Haematology, 50, 55-62.
COBURN, R.F., WILLIAMS. W . J . & K A H N . S . B . ( ~ ~ ~ ~ ) Endogenous carbon monoxide production in patients with hemolytic anemia. Journal of Clinical Investigation, 45, 4 6 0 4 6 8 .
CROSBY, W.H. (1 9 50) Hereditary non-spherocytic hemolytic anemia. Blood, 5, 233-241.
DACIE, J.V. & LEWIS, S.M. (1975) Practical Haema- tology, 5th edn. pp. 79-82. Churchill Living- stone, Edinburgh.
DANCEY. J.T.. DEUBELBEISS. K.A., HARKER, L.A. & GREENAWAY. J. (1976) Section preparation of human marrow for light microscopy. Journal of Clinical Pathology, 29, 704-710.
ENGSTEDT, L., JOHANSSON, S. & NYBERG. A. (1967) Estimation of human erythrocyte life-span from bilirubin turnover. Journal of Laboratory and Clinical Medicine, 70, 195-203.
EVERETT. B. (1974) Cluster Analysis, pp. 78-85. Heinemann Educational Books, London.
FINCH, C.A. (1982) Erythropoiesis, erythropoie- tin, and iron. Blood, 60, 1241-1246.
FINCH, C.A., DEUBELBEISS, K., COOK. J.D.. EISCH- BACH, J.W., HARKER, L.A., FUNK, D.D., MARSAG- LIA. G., HILLMAN, R.S.. SCHLICHTER, S., ADAM- SON, J.W.. GANZONI, A. & GIBLETT, E.K. (1 970) Ferrokinetics in man. Medicine, 49, 17-53.
GNANADESIKAN. R. (1977) Statistical Data Analysis of Multivariate Observations, pp. 13 7-19 5. John Wiley & Sons, New York.
GOWER, J.C. (1966) Some distance properties of latent root and vector methods in multivariate analysis. Biometrika, 53, 325-338.
HILLMAN, R.S. & FINCH, C.A. (1969) The misused reticulocyte. British Journal of Haematology, 17, 3 13-31 5.
INTERNATIONAL COMMITTEE FOR STANDARDIZATION IN HAEMATOLOGY ( 19 7 1 ) Recommended methods for radioisotope red cell survival studies. British Journal of Haematology, 21, 241-250.
3 70 G. Barosi et a1
INTERNATIONAL COMMITTEE FOR STANDARDIZATION I N HAEMATOL~CY ( I 9 78a) Recommended methods for measurements of serum iron in human blood. British journal lf Huemutology, 38, 291-294.
INTERNATION AI, COMMITTEE FOR STANDARDIZATION
IN HAEMATOLOCY (1 9 78b) The measurement of total and unsaturated iron-binding capacity in serum. British journal 01 Huc~matology, 38,
JENNRICH, K.I. ( 1 977) Stepwise discriminant analysis, Statistical Methods for Digital Com- putus (ed. by K. Enslein, A. Kalston and H. S. Wilf). pp. 76-95. John Wiley & Sons Ltd, New York.
LUNDH, B., CAVALLIN-STAHL, H. & MERCKE. C. (1975) Heme catabolism, carbon monoxide production and red cell survival in anaemia. Act0 Medica Srandinavica, 197, 141-171.
POOTRAKIJL, P.. HIJNCSPRENGES, S., FUCHAROEN, S., BAYLINK, D., THOMPSON, E., ENGLISH, E . . LEE, M..
2 8 1-290.
BURNELL, 1. & FINCH. C. (1981) Kelation between erythropoiesis and bone metabolism in thalassemia. New England journal of Medicine,
SAMSON, D.. HALLIDAY, D., NICHOLSON. D.C. & CHANARIN, I. (1 976) Quantitation of ineffective erythropoiesis from the incorportion of 1' 5N]delta-aminolaevulinic acid and [I 5N]gli- cine into early labelled bilirubin. f . Normal subjects, British lournal of Haematology. 34, 33-44.
STEPANELLI, M., BAROSI, G. & CAZZOLA, M. (1982) Iron kinetics. Quantitative Approarhm to Metab- olism (ed. by D. G. Cramp). pp. 143-183. John Wiley & Sons, Chichester.
WARD, H.P. & BLOCK, M.H. (1971) The natural history of agnogenic myeloid metaplasia (AMM) and a critical evaluation of its relation- ships with myeloproliferative syndrome. Medi- cine, 50, 357-420.