Serum selenium concentration at diagnosis and outcome in patients with haematological malignancies


Dr Simon Joel, Centre for Haemato-Oncology, Barts Cancer Institute, Queen Mary University of London, Charterhouse Square, London EC1M 6BQ, UK. E-mail:


We have previously reported presentation serum selenium level to be predictive of outcome in diffuse large B–cell lymphoma. This has now been studied in a further 430 patients, 163 with acute myeloid leukaemia (AML), 156 with Hodgkin Lymphoma (HL), and 111 with Follicular Lymphoma (FL). Serum selenium was below the UK normal reference range in 45% of patients, and correlated with serum albumin (r = 0·24–0·46, P < 0·001–0·003) in all tumour types. Independent predictors of presentation selenium were; French-American-British subtype and albumin (P < 0·001 for both) in AML, haemoglobin (P = 0·002) and B-symptoms (P = 0·01) in HL, and albumin (P < 0·001) in FL. In AML and HL, response to first line therapy was lower in patients with low serum selenium, but selenium was no longer predictive of response when other variables were entered into a multivariate model. Low selenium was also associated with a worse overall survival in FL [Hazard Ratio (HR) 2·3, 95% confidence interval (CI) 1·4, 4·0] and a trend to a worse overall survival in AML (HR 1·43, 95% CI 0·96, 2·13) by univariate Cox regression analysis, but not by multivariate analysis. In conclusion, low serum selenium is associated with a worse outcome in patients with haematological malignancies, but is not independently predictive, suggesting that it reflects other factors.

Selenium is a trace element essential to human health that is incorporated into at least 25 selenoproteins with a range of biological functions. Selenium deficiency is therefore associated with a range of disease states, including altered immune function, cardiac disease, inflammatory conditions and cancer. Over the last 30 years a number of studies have reported the cancer chemopreventive effects of selenium in animals [reviewed in (El-Bayoumy, 2001)] and in man (Clark et al, 1996), with the effect being clearest in prostate cancer (Duffield- Lillico et al, 2003), although a recently reported randomized trial failed to show a decreased hazard ratio (HR) for subsequent prostate cancer (1·04, 95% CI 0·87, 1·24) in healthy men receiving selenomethionine (Lippman et al, 2009).

We have previously reported serum selenium concentration at presentation to be independently predictive of both treatment response and long-term survival in patients with aggressive non-Hodgkin lymphoma (Last et al, 2003). Response to first treatment was only 54% in the lowest serum selenium quartile, compared to 88% in the highest quartile, with a lower overall survival in patients with lower serum selenium. Serum selenium remained predictive of outcome in a multivariate analysis that included clinical variables as cofactors. Similar observations have been made in cutaneous T-cell lymphoma patients (Deffuant et al, 1994).

Although these studies did not present a direct causal relationship between selenium and outcome, other reports have described an enhancement of the anti-cancer activity of chemotherapeutic agents when given with, or after selenium, but a reduction in toxicity to normal tissue (Cao et al, 2004; Juliger et al, 2007). Our own studies have shown selenium compounds to be cytotoxic as single agents in lymphoma cell lines (Last et al, 2006), but also to increase the sensitivity of these cells to cyclophosphamide, doxorubicin and etoposide at non-cytotoxic concentrations of selenium, possibly by reducing nuclear factor (NF)κB activity (Juliger et al, 2007). Additionally, in vivo xenograft studies have shown improved cure rates, but decreased toxicity, when organic selenium compounds were given before a number of cytotoxic drugs in colon and head and neck cancers (Cao et al, 2004). These data suggest that supranutritional doses of selenium may have a role clinically as an adjunct to chemotherapy.

There is little published data on presentation selenium and its association with outcome in other haematological malignancies. In children with leukaemia or lymphoma, hair selenium was lower than in a healthy control group (Ozgen et al, 2007), while serum selenium was decreased in adult patients with acute myeloid leukaemia (AML) (Beguin et al, 1986) or advanced chronic lymphocytic leukaemia (CLL) (Beguin et al, 1987). In this AML study, and in a more recent study, serum selenium normalized (Beguin et al, 1986), or increased (Asfour et al, 2009) in AML patients who achieved a complete remission.

We have therefore now extended our original study in diffuse large B-cell lymphoma to investigate the association between serum selenium at presentation and treatment and disease outcome in patients with AML, Hodgkin Lymphoma (HL) and Follicular Lymphoma (FL).

Patients and methods


Four hundred and thirty patients who presented to St Bartholomew’s Hospital between 1981 and 2005 were selected for inclusion in the study: 163 with AML, 111 with FL and 156 with HL. Selection was based on the availability of both a stored presentation serum sample in the tissue bank and clinical details in the patient database. All patients had given appropriate consent for the storage of their serum for research purposes. The study was approved by the East London and City Local Research Ethics Committee.

The diagnosis of AML prior to 2001 was made according to the French-American-British (FAB) classification (Bennett et al, 1976). Subsequently, diagnosis was based on the World Health Organization classification (Jaffe et al, 2001). Cytogenetic data was available on 140 out of 163 AML patients (86%). A histological diagnosis was made in all cases of HL and FL. Staging investigations comprised clinical examination, bone marrow aspirate and trephine biopsy and computed tomography scanning. All patients received standard induction therapy with curative intent, according to the protocols in use at the time.


For patients with AML, complete remission (CR) was defined as a normocellular bone marrow aspirate containing <5% blasts in the presence of normal maturation of all cell lineages. For patients with HL and FL, response to therapy was defined according to local criteria at the time; CR, good partial response (GPR), poor partial response (PPR) and no response [stable disease (SD) or progressive disease (PD)] (Johnson et al, 1995). GPR is comparable to CRu (unconfirmed) and PPR comparable to PR as defined in the consensus statement from the international working group (Johnson et al, 1995). Overall survival was measured as the time from the date of diagnosis to the date of last follow-up or death, event-free survival (EFS) in AML patients as the time from the date of diagnosis to the first of relapse, primary refractory disease or death, and time to transformation in FL patients from the date of diagnosis to the date of transformation, censored at death if no transformation had occurred.

Serum selenium determination

Total serum selenium was determined by inductively coupled plasma mass spectrometry for the principal isotope, selenium-78, after the addition of tellurium as internal standard (Delves & Sieniawska, 1997). The assay was standardized using matrix-matched standards containing bovine serum and quality controlled using certified reference material (Seronorm level 2; Nycomed Pharma AS, Oslo, Norway) and separate serum quality control samples (University of Surrey, Guildford, UK).

For statistical analysis serum selenium concentration at presentation was categorized as ‘low’ (<70·3 μg/l) or ‘normal’ (≥70·3 μg/l), according to the established laboratory normal reference range (70·3–157·9 μg/l).

Statistical methods

Associations between serum selenium and patient-specific variables were investigated using univariate and multivariate linear regression in the sas Statistics Package (SAS Institute Inc., Cary, NC, USA). Multivariate model building was by backward selection of variables with a P-value of <0·2 by univariate analysis based on minimizing the residual variance and number of parameters in the model.

Categorization of serum selenium into low or normal based on the reference range optimized the fit of the data to the proportional hazards assumption required for Cox regression. A maximum of one patient per disease had high (≥157·9 μg/l) selenium; these were grouped with ‘normal’.

Chi-squared tests for trend were used to assess the outcome of induction therapy according to selenium status (‘low’versus‘normal’), and logistic regression performed with response in two groups (these groups were defined differently for each disease). For logistic regression, model building and variable selection strategies were the same as for Cox modelling.

The effect of presentation selenium on EFS in AML, time to transformation in FL, and overall survival in each tumour type, was determined using the Kaplan–Meier method, (plotted using stata v10·1, StataCorp, College Station, TX, USA), and log-rank test where proportional hazards for selenium sub-groups was confirmed, and the Wilcoxon test where the hazards for sub-groups were not proportional. Multivariate Cox models for survival were also constructed, (variables with a P-value of <0·2 by univariate Cox analysis were included), with variable selection based on minimizing Akaike’s Information Criterion (AIC) value for the model. Variable selection was carried out before adding selenium to the model, so that the effect of selenium could be assessed in terms of what it added to the most informative model without selenium.


Presentation serum selenium

Patient characteristics and selenium concentrations for the three groups of patients (n = 430) are shown in Tables I and II. Overall, nearly half the patients entered into the study were in the low serum selenium group (<70·3 μg/l); 83 (51%) with AML, 29 (26%) with FL and 82 (53%) with HL. Only two of the 430 patients had a serum selenium concentration above the upper normal reference range (157·9 μg/l), one with FL and with HL.

Table I.   Patient characteristics and serum selenium concentration in patients with acute myeloid leukaemia (AML), Hodgkin Lymphoma (HL) and Follicular Lymphoma (FL).
 AML (= 163)HL (= 156)FL (= 111)
  1. Mean (SD) and median (range) are shown for normally distributed and non-normally distributed data respectively.

Serum selenium
 Mean, μg/l (SD)68·2 (24·2)70·2 (28·5)81·9 (22·3)
 Low (<70·3 μg/l)83 (51%)82 (53%)29 (26%)
 Normal (≥70·3 μg/l)80 (49%)74 (47%)82 (74%)
 Male96 (59%)99 (63%)52 (47%)
 Female67 (41%)57 (37%)59 (53%)
 Years, median (range)46 (16,60)30 (16,77)54 (24,87)
 g/l, mean (SD)38·6 (5·2) (n = 152)40·5 (5·4) (n = 151)42·8 (4·7) (n = 101)
 g/l, mean (SD)12·5 (2·1)13·3 (1·5)
White cell count
 1 × 109/l, median (range)9·2 (0·1, 448)  
B symptoms
 Present 70 (45%) 
 Absent 86 (55%) 
Stage or risk group I 37 (24%)I or Ie 21 (19%)
II 60 (38%)II or IIe 11 (10%)
III 29 (19%)III 20 (18%)
IV 30 (19%)IV 58 (52%)
 NA 1 (1%)
Table II.   acute myeloid leukaemia-specific characteristics.
FAB subtype n (%)Cytogenetic subtype n (%)
M01 (1)Good31 (19%)
M133 (20)Intermediate101 (62%)
M239 (24)Poor8 (5%)
M318 (11)Missing data23 (14%)
M425 (15)  
M524 (15)
M68 (5)
M72 (1)
Secondary AML10 (6)
Missing data3 (2)

There was a predominance of male patients in the AML and HL groups (59% and 63% respectively), in keeping with the known epidemiology of these diseases. The median age was 54 years in FL patients, but was much lower in HL patients (30 years). Only younger (≤60 years) AML patients were included in the study, resulting in a median age of 46 years. The majority of AML patients were in the intermediate cytogenetic risk group. Distribution of stage was as expected for HL and FL patients.

Factors influencing serum selenium concentration.  For each disease type the patient covariates were regressed against serum selenium concentration to identify associations between the two. These are summarized in Table III.

Table III.   Associations from univariate models between patient specific variables and serum selenium concentration.
Coeff95% CIr2 (P-value)Coeff95% CIr2 (P-value)Coeff95% CIr2 (P-value)
  1. r2 is a measure of goodness-of-fit and indicates the proportion of variation in selenium that is explained by the variable. Coeff, increase in selenium (μg/l); CI, confidence interval; FAB, French–American–British classification; AML, acute myeloid leukaemia; HL, Hodgkin Lymphoma; FL, Follicular Lymphoma.

Age (per decade increase)−0·38(−3·62, 2·86)0·0003 (0·82)1·99(−1·14, 5·13)0·01 (0·21)−3·34(−6·38, −0·29)0·04 (0·03)
Gender (Female versus Male)2·53(−5·10, 10·17)0·003 (0·51)0·03(−19·55, −1·09)0·03 (0·03)2·30(−6·12, 10·72)0·003 (0·59)
Albumin (per unit increase)1·49(0·80, 2·18)0·11 (<0·0001)0·06(0·43, 2·04)0·06 (0·003)2·16(1·32, 3·01)0·21 (<0·001)
Haemoglobin (per unit increase)0·08(1·82, 5·92)0·08 (0·0003)5·44(2·66, 8·22)0·13 (<0·001)
Stage or Risk Group (versus stage 1/ good prognosis)0·01 (0·38)0·090·09 (0·002)0·06 (0·08)
Intermediate: −6·48(−16·10, 3·13) (−29·90, −7·29) Stage 2: −6·40(−22·59, 9·79) 
Poor: −8·45(−27·01, 10·11) (−31·50, −4·68) Stage 3: −0·01(−13·61, 13·58) 
 (−36·84, −10·27) Stage 4: −11·71(−22·78, −0·63) 
White cell count (per 10-unit increase)−1·02(−1·57, −0·48)0·08 (0·0003)
B symptoms (versus none)0·06(−23·12, −5·54)0·06 (0·002)
FAB type (compared to M2)0·20 (<0·0001)
Secondary AML5·72(−10·02, 21·47)   
M044·44(−0·55, 89·42)   
M1−6·94(−17·45, 3·57)   
M39·26(−3·39, 21·92)   
M4−16·40(−27·78, −5·02)   
M5−18·44(−29·96, −6·91)   
M613·30(−3·94, 30·54)   
M7−0·66(−32·87, 79·68)   

For all three tumour types there was an association between albumin and serum selenium (AML and FL P < 0·001, HL P = 0·003). Other associations were seen with white cell count and FAB type in AML patients, haemoglobin in FL and stage, B symptoms and haemoglobin in HL.

Independent predictors of serum selenium were identified by backward selection of covariates with a P-value of <0·2 in the univariate analysis. In the AML group FAB type (P = 0·0008, effect size not given here as many subgroups) and albumin [coefficient 1·22 (95% confidence interval (CI) 0·51, 1·93); P = 0·0009] remained independent predictors, in HL haemoglobin (3·24 (1·17, 5·33); P = 0·002) and B symptoms [−11·04 (−19·85, −2·23); P = 0·01] were predictive, while in FL only albumin was predictive [2·16 (1·32, 3·01); P < 0·001].

Serum selenium and response to therapy

We next addressed whether presentation selenium was associated with response to first line therapy in each tumour type, as shown in Table IV and described below by tumour type.

Table IV.   The effect of presentation selenium on response to first line therapy in the three tumour types studied.
Response to therapyPresentation seleniumTotal
Acute myeloid leukaemia (χ2 test for trend P = 0·03)
 Complete remission48 (58%)57 (71%)105
 Resistant disease20 (24%)13 (16%)33
 Treatment-related death13 (16%)6 (8%)19
 Response unevaluable1 (1%)1 (1%)2
 No treatment1 (1%)3 (4%)4
Hodgkin lymphoma (χ2 test for trend P = 0·04)
 Complete response31 (38%)43 (58%)74
 Good partial response28 (34%)19 (26%)47
 Poor partial response12 (15%)6 (8%)18
 Stable disease4 (5%)1 (1%)5
 Progressive disease7 (8%)5 (7%)12
Follicular lymphoma (χ2 test for trend P = 0·37)
 Complete response7 (24%)32 (39%)39
 Good partial response9 (31%)17 (21%)26
 Poor partial response5 (17%)14 (17%)19
 Stable disease0 (0%)4 (5%)4
 Progressive disease2 (7%)5 (6%)7
 Death1 (3%)0 (0%)1
 Response unevaluable5 (17%)10 (12%)15

AML.  A Chi-squared test for trend showed a significantly worse response to therapy in AML patients who had low selenium status at presentation (P = 0·03). This was due to a lower remission rate and an increased incidence of both resistant disease and treatment-related death. On fitting a univariate regression model for association of selenium (low versus normal) and response to induction therapy in AML (CR versus all death and resistant disease grouped), low selenium was associated with a worse response to induction therapy [odds ratio (OR) 0·38, 95% CI 0·17, 0·85 P = 0·02]. On multivariate analysis, after the inclusion of cytogenetic risk group [OR intermediate versus good 0·21 (0·06, 0·78), OR poor versus good 0·16 (0·02, 1·30); P = 0·06] and albumin [OR per unit increase 1·10 (1·02, 1·20); P = 0·03] to the model, serum selenium was no longer significant [low versus normal OR 0·59 (0·25, 1·42); P = 0·24].

HL.  A Chi-squared test for trend also showed a significantly worse response to therapy in HL patients with low presentation selenium (P = 0·04), due mainly to a superior complete response rate in patients with normal selenium. Consequently if CR and GPR are grouped together, as commonly done when summarizing responses in HL, compared with all other outcomes (PPR, SD and PD), the effect of selenium is no longer significant (Chi-squared test for trend P = 0·08). By univariate regression analysis, presentation selenium was also associated with response to therapy, with an increase of 10 μg/l in the serum selenium concentration increasing the OR for a CR to 1·17 (95% CI 1·03, 1·31, P = 0·01). However, the most significant predictor of achieving a CR (versus other responses) was stage [OR per 1-stage increase 0·34 (0·23, 0·52); P < 0·001], such that once stage was entered into a multivariate model selenium was no longer predictive [OR per 10 μg/l increase 1·09 (0·95, 1·24); P = 0·22].

FL.  Presentation selenium was not associated with response to therapy in patients with FL (P = 0·37).

The effect of presentation serum selenium on outcome

Serum selenium and overall survival.  The effect of selenium (low versus normal) on EFS (AML), time to transformation (FL), and overall survival was investigated. Kaplan–Meier overall survival plots for each group are shown in Fig 1.

Figure 1.

 Overall survival curve (Kaplan–Meier) by presentation selenium (low versus normal) in patients with (A) AML, (B) HL and (C) FL.

AML.  Presentation selenium was not associated with EFS (logrank P = 0·18) in AML patients, but was associated with overall survival, although the effect did not reach statistical significance (logrank P = 0·08), with an early separation in the survival curves not maintained throughout the period of follow-up. Univariate Cox regression analyzes gave a HR for patients with low selenium of 1·43 (95% CI 0·96 to 2·13, P = 0·08), suggesting poorer overall survival in patients with low selenium. Multivariate Cox modelling (including age, albumin, risk group and white cell count) was then used to identify independent predictors of overall survival. Cytogenetic risk group [HR Intermediate versus good 2·58 (1·35, 4·93), HR poor versus good 3·43 (1·24, 9·47); P = 0·01] and albumin [HR per unit increase 0·92 (0·89, 0·96); P = 0·0003] were independently associated with overall survival, but white cell count (P = 0·26) and age (P = 0·18) were not. Selenium was no longer significant after adjusting for risk group and albumin [HR 1·03 (95% CI 0·63, 1·70); P = 0·9]. Because of the strong interaction between serum selenium at presentation and serum albumin, an interaction term was introduced into the Cox multivariate analysis. In this model the interaction was predictive of overall survival (P = 0·05). This suggested a complex relationship between the two variables and outcome: if the albumin was high, patients with a normal serum selenium had a better outcome, while if the albumin was low the converse was true, with the low selenium group having a better overall survival. This relationship is presented graphically in Fig 2.

Figure 2.

 Change in hazard ratio for varying serum albumin concentration in patients with low (<70·3 μg/l) or normal (≥70·3 μg/l) serum selenium. Hazard ratios are in comparison to baseline hazard: albumin of 40 g/l, normal selenium.

HL.  As with AML patients, presentation selenium in HL patients was associated with overall survival (Wilcoxon P = 0·05 for HL), with a quite marked early separation in the survival curves not maintained throughout the follow-up period. It was noticeable that deaths in the HL patient group with low serum selenium were in the first 6 years of follow-up, with no deaths after that time-point.

Cox modelling was not carried out in HL patients due to non-proportional hazards for the selenium groups.

FL.  In FL there was a marked difference in overall survival between patients with low and normal selenium (logrank P = 0·002, Fig 1C), with univariate Cox regression analysis giving a HR for patients with low selenium of 2·3 (95% confidence interval 1·4–4·0, P = 0·002) Multivariate Cox modelling (including age, albumin, stage, gender and haemoglobin) showed age [HR per decade increase 2·29 (1·78, 2·96); P < 0·001], stage [HR per 1-stage increase 1·74 (1·33, 2·29); P < 0·001], and gender [HR female versus male 0·52 (0·30, 0·89); P = 0·02] to be predictive of outcome, but haemoglobin (P = 0·18) and albumin (P = 0·46) were not. Selenium showed borderline significance after adjusting for age, stage and gender (HR 1·7, 95% CI 0·98–3·0, P = 0·06). Presentation serum selenium showed no association with time to transformation (logrank P = 0·92).


This is the largest study to date that has investigated the influence of serum selenium at presentation on outcome in patients with haematological malignancies. Over half of the patients in the study with newly diagnosed AML or HL, and 26% of those with FL, had a selenium concentration that would be considered low in the UK. The low serum selenium in these patients confirmed the same observation in our previous study in aggressive B-cell lymphomas (Last et al, 2003) and other smaller studies in CLL (Beguin et al, 1987), AML (Beguin et al, 1989) and lymphoma (Avanzini et al, 1995).

The reasons for the low selenium status in these haematological malignancies are not clear, although several possibilities can be considered.

The first is that selenium deficiency is due to inadequate dietary intake. Although some patients presenting with AML may have anorexia, nausea or vomiting, the course of the acute illness is usually only a matter of weeks, likely to be insufficient to result in severe systemic deficiency. In contrast, HL patients have a more sub-acute presentation, and nearly 50% of the HL group presented with B symptoms, one feature of which is a > 10% weight loss over the preceding 6 month period. B-symptoms are less commonly seen in FL patients [typically around 20% at presentation (van Besien et al, 1998)], as was decreased serum selenium.

A second possibility is that low serum selenium is a function of the malignant process, resulting from the increased utilization of selenium by leukaemia or lymphoma cells. This possibility is supported by a study in women with breast cancer, in whom low serum selenium concentrations were correlated with increased concentrations of selenium in both malignant and adjacent benign breast tissue, suggesting that malignant tissue has a higher uptake of selenium (Charalabopoulos et al, 2006). In our own study there was a significant negative correlation between white cell count and selenium concentration (P = 0·0003) in AML patients, and selenium and disease stage in HL (P = 0·002). Other reports have described a similar negative correlation between serum selenium and lymphocyte count in CLL (Beguin et al, 1987) peripheral blast cell count in AML (Beguin et al, 1989) and disease stage in lymphomas (Avanzini et al, 1995), suggesting that utilization of selenium by the tumour may decrease serum selenium levels and act as a surrogate for bulk of disease.

The third and most likely possibility is that low selenium concentration in these malignancies is part of a systemic, acute phase response caused by the disease. Our own data lends weight to this hypothesis. Serum albumin is regarded as an important marker of the acute phase response, decreasing as a response to inflammation due to the inhibitory effect of cytokines and tumour necrosis factor-α on ALB mRNA transcription, resulting in a decrease in the synthesis of albumin by the liver (Brenner et al, 1990). On univariate analysis serum selenium was positively associated with serum albumin in all three tumour types, and remained significant in AML and FL in multivariate analysis. In HL patients the only factors independently predictive of serum selenium concentration at presentation were B symptoms and haemoglobin, with albumin not significant once B symptoms was entered into the model.

Other published data support the relationship between serum selenium and an acute phase response. Patients admitted to an intensive care unit with critical illness have been reported to have low serum selenium concentration, which correlated negatively with the Acute Physiology and Chronic Health Evaluation Score (APACHE II), a validated scoring system for assessing the severity of illness for critically ill intensive care patients (Forceville et al, 1998). The mortality rate was also 3·5 times higher in the group with the lowest serum selenium concentration. Decreased serum selenium is also seen in patients presenting with increased C-reactive protein, rheumatoid arthritis, lung cancer and T-cell acute lymphoblastic leukaemia (Maehira et al, 2002). A possible mechanism for this decrease in serum selenium in patients with an acute phase response may be a decrease in selenoprotein P, as observed in patients with sepsis (Hollenbach et al, 2008), Crohn’s disease (Andoh et al, 2005) and recently in patients with prostate cancer (Meyer et al, 2009). Selenoprotein P is the main selenoprotein in serum, accounting for 40–50% of total serum selenium (Moschos, 2000). A study in mice has related the acute phase response to a decrease in selenoprotein P translation in the liver, which in turn affected selenium metabolism and transport (Renko et al, 2009). This relationship is made more complex by the fact that selenoprotein P is itself a sensitive indicator of selenium status with a very good correlation between serum selenium and selenoprotein P concentration (Moschos, 2000). Polymorphisms in selenoprotein P have also been reported to influence serum selenium concentration (Meplan et al, 2007).

We next investigated whether presentation selenium (low or normal) was associated with response to treatment or survival. Serum selenium was associated with response to therapy using univariate analysis in patients with AML (P = 0·03) and HL (P = 0·04), but not in those with FL (P = 0·37). However, in both AML and HL selenium was no longer predictive of response when other covariates were added, specifically cytogenetic risk group and albumin in AML and stage in HL.

The clearest association between serum selenium and survival was in FL, with a logrank P-value of 0·002 and univariate Cox HR for a patient in the low selenium group of 2·3 (CI 1·4, 4·0, P = 0·002). This survival effect was not attributable to a difference in the time to transformation to DLBCL, which was not different between the low and normal selenium groups (logrank P = 0·92). In AML and HL patients there was a less clear association between presentation selenium and survival, with a logrank p value of 0·08 for AML and Wilcoxon (non-proportional hazards) P value of 0·05 for HL patients. In AML patients there was no difference in progression-free survival between those with low and normal selenium, suggesting that once CR is achieved the effect of selenium status at presentation is lost, with a high incidence of relapse in both groups. In HL, patients with selenium levels below the normal range were more likely to die in the first few years after diagnosis, but there were more later deaths in the normal selenium group. These effects balance out to leave comparable survival rates after 15–20 years in both groups, (74% vs. 77% in low versus normal selenium groups, respectively, at 15 years and 74% in each group at 20 years).

On multivariate analysis of these data the effect of selenium as a predictor of survival was reduced. In FL, multivariate Cox modelling found age, stage and gender to be independently predictive of survival, with presentation selenium of borderline significance [P = 0·06, HR 1·7 (0·98, 3·0)]. In AML the inclusion of cytogenetic risk group and albumin dramatically reduced the importance of serum selenium (P = 0·9). As there was a strong univariate relationship between selenium and albumin in AML, an interaction term was introduced into the multivariate model that restored the importance of selenium as a predictor of survival (P = 0·05). In this interaction, patients with low selenium had a worse outcome in the presence of a normal albumin, but not when the albumin was low. This effect may be attributable to the fact that some selenium in serum is bound to albumin, such that although total serum selenium would be expected to be lower in patients with low albumin, because of the reduced availability of binding sites the free fraction may increase. This may also reflect the importance of the acute phase response, likely reflected as a decreased albumin in these patients, in which serum selenium, although low (<70·3 μg/l), is no longer predictive of outcome. However, this interaction was not a pre-specified analysis and its significance may be a chance finding not repeatable in other data sets.

Overall, the data presented here suggest that serum selenium is associated with outcome in each of the malignancies studied using univariate analyzes, but is also strongly associated with other variables, such as albumin and stage. The strong association between these other factors and outcome means that when they are also included in outcome analyzes serum selenium at presentation is no longer independently predictive of overall survival, or response to first therapy. It is therefore not possible from these data to confirm or refute whether there is a direct causal relationship between serum selenium and outcome in these diseases; an interventional study is required for this.

There are similarities and differences between these data and that from our previous study in patients with diffuse large B-cell lymphoma (Last et al, 2003). In that study, serum selenium at presentation correlated closely with only one other clinical variable, performance status, possibly again reflecting the relationship between serum selenium and nutritional deficiency or the presence of an acute phase response. The dependent nature of these factors was evident in the multivariate analyzes; when either performance status or selenium was entered first the other was no longer predictive. Excluding performance status from these analyzes, presentation selenium remained predictive of dose delivery, response to treatment and overall survival.

Our own group, and others, have reported that supra-nutritional doses of selenium compounds, that elevate serum selenium to well above the normal range, do have chemomodulatory activity (Ip et al, 2000), or affect outcome in the cancer setting. Our in vitro studies have shown that non-toxic concentrations of organic selenium increased the apoptotic activity of several established anti-lymphoma agents, possibly via an effect on NFkB activity (Juliger et al, 2007). In a xenograft model of lung and head and neck cancer, dosing of organic selenium species before and during chemotherapy, thereby raising the serum selenium level to markedly higher than the normal range, improved the efficacy (cure rate), but also decreased the toxicity of a number of cytotoxic drugs (Cao et al, 2004). Finally, in a clinical trial of selenium supplementation with cisplatin-based chemotherapy, patients were reported to experience reduced haematological toxicity and nephrotoxicity (Hu et al, 1997). In another disease setting, intensive care patients randomized to receive 1000 μg of sodium selenite had a significant reduction in mortality rate compared to those who received placebo, particularly so for patients with septic shock with disseminated intravascular coagulation, the most critically ill patients, or patients with more than three organ dysfunctions (Angstwurm et al, 2007). No selenium-induced toxicity was observed.

In summary, low serum selenium at presentation is associated with a worse outcome in patients with AML, FL and HL, but is not independently predictive of survival or response to treatment. It is likely that low serum selenium in such patients reflects ongoing nutritional deficiency, an acute phase response and/or extent of disease.