Background

The prevalence of anaemia, a reduction in the number of red blood cells, in older individuals is around 17% and increases with age to above 20% in those aged 85+ years [1]. Older persons with anaemia often report poor quality of life and complaints of fatigue and weakness. Anaemia in old age has been associated with a higher risk of falls, frailty, disability, depression, longer postoperative recovery, as well as lower physical and cognitive capacity, a higher rate of comorbidity, and an increased mortality risk [2,3,4,5,6,7,8].

Since anaemia is a highly prevalent and incident disorder in older persons with relevant consequences on functioning and survival, previous research has focussed on single causes of anaemia in old age, such as nutritional deficiencies (iron, vitamin B12, and folate deficiency) [9, 10], renal impairment and chronic inflammation [11,12,13].

Previous findings have shown that the combinations of determinants are associated with poor outcomes, such as multimorbidity and in-hospital mortality [14], and polypharmacy and drug-disease interactions [15]. In addition, laboratory profiles have shown an association with all-cause mortality risk in old age with increasing risk per laboratory abnormality [16]. We therefore hypothesised that an increment of determinants of anaemia increases the possibility (risk) of anaemia in old age. We investigated the effects of single, and a combination of determinants on anaemia among the very old cross-sectionally and prospectively in the TULIPS Consortium, a collaboration of five international cohorts from four studies in very old individuals.

Methods

Study population and procedures

The Towards Understanding Longitudinal International Older People Studies (TULIPS) Consortium was established in 2014 and is a collaboration of five cohorts from four studies in four countries: Leiden 85-plus Study in the Netherlands, LiLACS NZ in New Zealand, Newcastle 85+ study in the United Kingdom, and TOOTH in Japan. All are observational, prospective and population-based studies that aim to investigate consequences of health and disease determinants in very old individuals.

The Leiden 85-plus Study recruited 85-year-old inhabitants in Leiden, the Netherlands. From September 1997 through September 1999, 705 inhabitants reached the age of 85 years and were eligible for participation in the study. Subjects were visited annually (at ages 85–90 years for 5 years) at their place of residence for face-to-face interviews and cognitive tests. Blood samples were taken according to predefined protocols under non-fasting conditions. The medical history was obtained from the participants, or partners and caregivers in memory-impaired subjects. Pharmacy records were studied annually, and participants were also interviewed to evaluate the use of (over-the-counter) supplements. The Medical Ethical Committee of the Leiden University Medical Center approved the study, and informed consent was obtained from all subjects or their guardian for cognitively impaired subjects [17].

The Life and Living in Advanced Age (LiLACS NZ) is a population-based study of two cohorts: Māori and non-Māori in advanced age in New Zealand. Participants were living in the Bay of Plenty and were born between 1 January, 1925 and 31 December, 1925 for non-Māori (85 years) and between 1920 and 1930 for Māori (80–90 years old) [18, 19]. At baseline, participants were interviewed using a comprehensive questionnaire, conducted by trained interviewers, and had a health assessment conducted by a trained nurse. Blood samples were collected at baseline after an overnight fasting. Information about medication and supplement intake was collected. A simplified version of the questionnaire was made available for those unable to do the comprehensive one [18]. Ethical approval for the LiLACS NZ was granted by the Northern X Regional Ethics Committee (NTX 09/09/088) in December 2009 and written informed consent was obtained from all participants for each stage of the study [19].

The Newcastle 85+ study recruited individuals turning 85 years old in 2006 (born in 1921) and who were registered with a participating National Health Service (NHS) general practice in the areas of Newcastle upon Tyne or North Tyneside in the north-east of England. Older persons living in care homes and with disabilities and/or cognitive impairment were included. At baseline (2006/2007), participants underwent a detailed multidimensional health assessment by trained research nurses in their residence (own home or institution), containing questionnaires, measurements, function tests, a fasting blood sample, and a review of medical records held by the general practice [20, 21]. Participants were followed up for an assessment of health and functioning (including self-reported physical activity) at 18 months (1.5 years; wave 2), 36 (3 years; wave 3), and 60 months (5 years; wave 4) by research nurses at their usual place of residence [22, 23]. For the present analysis, data were available for wave 2 and wave 3. The study was approved by the Newcastle and North Tyneside local research ethics committee (06/Q0905/2). Written informed consent was obtained from all participants, otherwise from a caregiver or a relative for those with cognitive impairments [20].

The Tokyo Oldest Old Survey on Total Health (TOOTH) recruited a random sample of community-dwelling seniors aged 85 years or older in the city of Tokyo, Japan. The baseline comprehensive assessment consisted of an in-home interview, a self-administered questionnaire, and a medical/dental examination. Participants were invited to Keio University Hospital for biomedical assessments. For those who were unable to visit the study centre, home-based examinations were done through visits by a geriatrician and a dentist. Socio-demographic characteristics and diseases at baseline were collected. However, since not all determinants were collected in the first year in TOOTH, we defined 3-year follow-up as baseline data, and 6-year follow-up as follow-up variable. The study was approved by the ethical committee of the Keio University School of Medicine (N0. 20,070,047, Dec2007) [24]. Written informed consent was obtained either from the participants or by their proxy (normally a family member or caregiver) if necessary.

A flow chart of recruitment and schematic representation of data samples in the five cohorts is shown in Supplementary Figure 1.

Study parameters/measures

Outcome

In all studies, haemoglobin was measured in anti-coagulated whole blood samples (ethylenediaminetetraacetic acid (EDTA)-coated tubes) using standard automated analysis systems. The methods and analysers are listed in Supplementary Table 2. Anaemia was defined by the World Health Organization (WHO) as haemoglobin < 13 g/dL in men and < 12 g/dL in women [25]. In the cross-sectional analyses, the presence of anaemia at baseline served as the outcome variable; in the prospective analyses, the onset of anaemia during follow-up was the outcome variable. When a new case of anaemia was identified, we assumed that the anaemia had developed halfway the period between two follow-up blood measurements.

Determinants of anaemia

Iron deficiency, vitamin B12 deficiency, folate deficiency, low estimated glomerular filtration rate (eGFR), and high C-reactive protein (CRP) were selected as single determinants of anaemia and were analysed to investigate their association with the presence and onset of anaemia.

Ferritin level was considered an indicator of iron status [26], with the cut-off of < 15 μg/L for women and < 20 μg/L for men as deficiency. Vitamin B12 deficiency was defined as < 150 pmol/L. [27, 28] Folate deficiency was defined using specific cut-offs in the five cohorts, which were serum folate level < 7 nmol/L for the Leiden 85-plus Study and TOOTH [28], red blood cell folate < 317 nmol/L for LiLACS NZ [19] and red blood cell folate < 340 nmol/L for the Newcastle 85+ study [21, 29]. The MDRD (Modification of Diet in Renal Disease) Study equation was adopted to calculate eGFR [30,31,32]. Low eGFR was defined as eGFR below 45 ml/min/1.73m2 [12]. C-reactive protein levels higher than 5 mg/L were defined as high CRP [33].

Determinants with abnormal values were defined as 1 and those within normal range as 0. The combination of determinants was calculated as the sum of abnormal determinants at baseline ranging 0–5. Participants with missing data on single determinants were considered as within the reference intervals (In total, 104 (4.7%) for ferritin; 161 (7.3%) for vitamin B12; 103 (4.6%) for folate; 4 (0.2%) for eGFR; 6 (0.3%) for CRP). Those with missing data of all determinants were excluded from the analyses (n = 4). Then, participants were divided into two groups based on the sum of the determinants: 0 to 1 vs. 2 and above determinants.

Other variables

As socio-demographic characteristics, we selected age, sex and institutionalisation, which were available in the five cohorts. Institutionalisation was regarded as living in a nursing home. Since only community-dwelling older adults were recruited in TOOTH, this variable was not applicable. Smoking was classified as non-smokers and past or current smokers. Multi-morbidity was composed of the presence or history of cerebrovascular accident (stroke), coronary heart disease (CHD), cancer and diabetes. It was stratified into 0 to 1 versus ≥2 as a binary variable in the analyses.

Statistical analyses

First, each cohort was analysed cross-sectionally to investigate the association of single and a combination of determinants with the presence of anaemia. Then, the studies were analysed prospectively to investigate the effects of the single and the combination of the determinants on the development of incident anaemia. After analysing the studies separately, the data were pooled in a meta-analysis. We further explored the effects of using a cut-off of ferritin of 50 μg/L for iron deficiency, and a combination of significant single determinants as a sensitivity analysis.

Multivariate logistic regression analyses were used to calculate odds ratios (OR) and their 95% confidence intervals. We applied three models: model 1) unadjusted model; model 2) adjusted for socio-demographic characteristics (age, sex, institutionalisation) and smoking; and model 3) a fully adjusted model which included socio-demographic characteristics, smoking and multi-morbidity.

Prospective analyses were done in three studies for which follow-up data were available: the Leiden 85-plus Study, the Newcastle 85+ study and TOOTH. Participants with anaemia at baseline were excluded for the prospective analyses. Multivariate Cox Proportional-Hazards Models were performed, with adjustments for the aforementioned confounding variables.

After acquiring the individual cross-sectional results of the five cohorts and prospective results of the three studies, they were pooled using a two-stage individual participant data (IPD) meta-analysis [34]. Because of the clinical heterogeneity between the studies, random-effects models were applied.

The statistical analyses were performed with SPSS software (version 25) and Review Manager (RevMan). Version 5.3. Copenhagen: The Nordic Cochrane Centre, The Cochrane Collaboration, 2014. for meta-analyses.

Results

Table 1 shows the baseline characteristics of the five cohorts. All participants in the Leiden 85-plus Study and the Newcastle 85+ study were 85 years old. The median age of the participants in the LiLACS NZ was 82 years for Māori and 85 years for non-Māori; and 90 years for TOOTH. Overall, among a total of 2216 participants, 59% were female and 30% had anaemia at baseline (21.0% for non-Māori, 21.7% for Māori in the LiLACS NZ, 28.5% in the Leiden 85-plus Study, 29.8% in the Newcastle 85+ study and 48.4% in TOOTH). Furthermore, subjects with a combination of ≥2 determinants accounted for 4.6% in TOOTH, 21.8% in Leiden 85-plus Study, 21.8% in the Newcastle 85+ study and 24.1% for non-Māori and 24.6% for Māori in the LiLACS NZ.

Table 1 Baseline Characteristics of Participants in Four Population-Based Studies of the Very-oldab
Full size table

Cross-sectional results per single determinant

Supplementary Table 3 shows the prevalence of anaemia at baseline, depending on the presence of single and a combination of determinants.

Detailed results of unadjusted models and the two models with adjustments for confounders are displayed in Fig. 1 and Supplementary Table 4. After pooling the fully adjusted results of the five cohorts, the pooled adjusted OR for the presence of anaemia for participants with iron deficiency compared to their counterparts without iron deficiency was 2.76 (95% CI, 1.87 to 4.07, I2 = 24%). There was no evidence of an association between either vitamin B12 deficiency or folate deficiency and the presence of anaemia. Low eGFR and high CRP were associated with the presence of anaemia.

Fig. 1
figure 1

Cross-sectional Results: Meta-Analyses of Determinants of Anaemia at Baseline in Association with Presence of Anaemia abc. Abbreviations: OR, odds ratio; CI, confidence interval; eGFR, estimated glomerular filtration rate; CRP, C-reactive protein. a Iron deficiency was defined as ferritin < 20 μg/L for men, < 15 μg/L for women; vitamin B12 deficiency was < 150 pmol/L; folate deficiency was serum folate level < 7 nmol/L (Leiden 85-plus Study and TOOTH) or red blood cell folate < 317 nmol/L (LiLACS NZ) and < 340 nmol/L (Newcastle 85+ study); low eGFR was < 45 mL/min/1.73 m2, eGFR was calculated using MDRD (Modification of Diet in Renal Disease) Study equation from the National Kidney Foundation; high CRP was > 5 mg/L. b Results of fully adjusted model (model 3): adjusted for age, sex, institutionalisation, smoking and ≥ 2 multi-morbidity. Multi-morbidity was composed of stroke, coronary heart disease (CHD), cancer and diabetes. It was stratified into 0 to 1 or 2 and above as a binary variable. Leiden 85-plus Study: sex, institutionalisation, smoking and ≥ 2 multi-morbidity [stroke, coronary heart disease (CHD) excluding stroke, cancer, diabetes]; LiLACS NZ: age, sex, institutionalisation, smoking and ≥ 2 multi-morbidity [stroke (cerebrovascular accident (CVA), cardiovascular disease (CVD) excluding stroke, cancer, diabetes]; Newcastle 85+ study: age, sex, institutionalisation, smoking, ≥2 multi-morbidity (CVA, combined cardiac disease excluding CVA, cancer, diabetes); TOOTH: age, sex, smoking, ≥2 multi-morbidity (stroke, coronary heart disease (CHD), cancer, diabetes). c LiLACS NZ contained two cohorts: Māori and non-Māori population; TOOTH: since not all determinants were collected at baseline, 3-year follow-up was defined as baseline, and 6-year follow-up as follow-up data. d All four studies included five determinants: iron, vitamin B12, folate deficiency, low eGFR, and high CRP. e Population with determinant within total anemic population. f Population with determinant within total non-anemic population

Full size image

Cross-sectional results for the combination of determinants

The prevalence of 0, 1, 2, 3, 4 and 5 determinants for participants with and without anaemia are shown in Supplementary Table 5. Overall, between 17.8 and 56.9% of the participants with anaemia had a sumscore of 0. Per additional determinant the risk of the presence of anaemia was 2.10 (95% CI, 1.85 to 2.38, I2 = 0%, P trend< 0.001) (Figs. 1, 2). In addition, a combination of two or more abnormal determinants was significantly associated with the presence of anaemia in all the five cohorts, with a pooled adjusted OR of 3.44 (95% CI, 2.70 to 4.38, I2 = 0%; Fig. 1).

Fig. 2
figure 2

Cross-sectional Results: Trends of Odds Ratios of Combination of Determinants on the Presence of Anaemia ab. Abbreviations: OR, odds ratio; CI, confidence interval. Participants with 0 determinant were the reference group. a Very low number of population size in some subgroups led to an inestimable odds ratio. b Results of fully adjusted model (model 3): adjusted for age, sex, institutionalisation, smoking and ≥ 2 multi-morbidity

Full size image

Prospective results per single determinant

Supplementary Table 6 shows the incidence of anaemia at baseline, depending on the presence of single and a combination of determinants.

In the pooled analyses, participants with iron deficiency (Table 2), vitamin B12 deficiency, or folate deficiency (Table 2) did not have a higher risk of developing anaemia during follow-up. Low eGFR was associated with the onset of anaemia (pooled adjusted HR 1.62 95% CI, 1.04 to 2.54, I2 = 35%; Table 2). In addition, the pooled estimate of high CRP on the onset of anaemia was 1.40 (95% CI, 1.04 to 1.89, I2 = 0%; Table 2).

Table 2 Meta-Analyses of Single and Combination of Determinants of Anaemia in Association with Onset of Anaemia abc
Full size table

Prospective results for the combination of determinants

Per additional determinant the risk of developing anaemia was statistically significant in the Leiden 85-plus Study (HR 1.35 (95% CI, 1.06 to 1.73) and the Newcastle 85+ study (HR 1.58 (95% CI, 1.25 to 1.98) (Table 2), but not for TOOTH. The pooled estimate of the effect on the development of anaemia was 1.46 (95% CI, 1.24 to 1.71, I2 = 0%, P trend< 0.001) per additional determinant (Table 2, Fig. 3).

Fig. 3
figure 3

Prospective Results: Trends of Hazard Ratios of Combination of Determinants on the Onset of Anaemia ab. Abbreviations: HR, hazard ratio; CI, confidence interval. Participants with 0 determinants were the reference group. a Very low number of population size in some subgroups led to an inestimable odds ratio. b Results of fully adjusted model (model 3): adjusted for age, sex, institutionalisation, smoking and ≥ 2 multi-morbidity

Full size image

Participants with a combination of ≥2 determinants had a statistically significant 1.87-fold higher risk of developing anaemia compared to those with 0 or 1 determinant (95% CI, 1.12 to 3.12) in the Leiden 85-plus Study and Newcastle 85+ study (HR (95% CI), 2.02 (1.30 to 3.13)), but not in TOOTH. The pooled effect estimate for the onset of anaemia was 1.95 (95% CI, 1.40 to 2.7165, I2 = 0%; Table 2, Supplementary Table 7).

Sensitivity analyses

Similar results were obtained when a cut-off of 50 μg/L was used to define iron deficiency (Supplementary Tables 8, 9, 10, 11 and 12).

We further explored the effects of a combination of single determinants (iron deficiency, low eGFR and high CRP) that were individually associated with the presence or onset of anaemia, as a sensitivity analysis.

In the cross-sectional analyses, the pooled estimate for the OR on the presence of anaemia was 2.64 (95% CI, 2.22 to 3.14, I2 = 15%) per additional determinant. In addition, compared to participants with 0 or 1 determinant, the pooled adjusted OR for a combination of ≥2 of the three significant determinants (iron deficiency, low eGFR and high CRP) on the presence of anaemia was 4.53 (95% CI, 2.66 to 7.72, I2 = 50%).

In the prospective analyses, per additional determinant, the risk of development of anaemia increased 1.54 fold (pooled adjusted HR (95% CI), 1.54 (1.24 to 1.92), I2 = 11%). In addition, participants with a combination of ≥2 significant determinants had a 1.99-fold higher risk of developing anaemia compared to those with 0 or 1 determinant (pooled adjusted HR 1.99 95% CI, 1.32 to 3.02, I2 = 0%).

Discussion

Principal findings

In this study, we found that very old adults with iron deficiency, low eGFR and high CRP at baseline had a higher risk of having anaemia. A combination of two or more determinants at baseline was associated with the presence of anaemia. Moreover, low eGFR, high CRP and a combination of two and above determinants were associated with the development of anaemia during follow-up.

Comparison with other studies

This study confirmed the well-known association of iron deficiency and the presence of anaemia [35]. Surprisingly, iron deficiency was not associated with anaemia prospectively [36, 37]. Comparable with previous studies in other settings, low eGFR and high CRP were also associated with both the presence and the development of anaemia in this study [38,39,40].

We found an increased risk of folate deficiency on anaemia in the Leiden 85-plus Study that corresponded to the previous conclusion in 2008 [27]. However, in the pooled analysis in the TULIPS Consortium, we found no evidence that vitamin B12 and folate deficiency were associated with the presence or onset of anaemia. This is in line with earlier systematic reviews and meta-analyses showing no effect of treatment with vitamin B12 or folic acid on haematological parameters in older individuals [28].

The prevalence of anaemia in TOOTH was almost double that of the other four cohorts but the percentages of single determinants were half. This is in accordance to previous research stating that the frequency of anaemia in Japan is particularly high compared to other countries; moreover, other causes of anaemia such as myelodysplastic syndromes in Japan may be an important determinant for future studies [41].

Strengths

First, having access to original datasets enabled us to conduct an individual participant data (IPD) meta-analysis to standardize definitions and analyses. In addition, the two-stage method allowed us present the individual study data and to explicitly show the differences in cohorts, and apply well-described standard meta-analysis methods [34]. Second, in the TULIPS Consortium, five cohorts in the very old population were combined which resulted in a large number of very old participants from across the world. This means our results can be extrapolated to a large and diverse group of older individuals worldwide.

Weaknesses & limitations

We recognize several limitations in our study. First, no full diagnostic work-up for anaemia was performed. Still, our primary analyses are based on measurements of iron status, vitamin B12, folate, CRP and creatinine, which are direct measures or surrogates for the most commonly mentioned causes of anaemia in older persons [1]. These measures are also used in clinical practice to further investigate the cause of anaemia. In addition, we incorporated major clinical conditions such as cardiovascular disease and cancer as important indicators of clinical status, which we accounted for in the analyses. Second, information on serum iron, iron saturation, MCV or other blood counts were not available in all studies. Third, we adjusted for a limited number of comorbidities (cardiovascular disease, cancer and diabetes), as information on other comorbidities was not present in all studies. Residual confounding by e.g. liver disease or gastro-intestinal bleeding may have led to an overestimation of the true associations. Fourth different methods and analysers were applied for the laboratory measurements in the five cohorts. However, the laboratory methods were consistent within each study and were analysed on a study level, which was likely not to have resulted in systematic error, making the effect of differences in laboratory measurements a minor concern. Fifth, small numbers in some subgroups led to inestimable results which could not be pooled in the meta-analyses.

In addition, participants with missing data on single determinants were assumed to be within normal range which could have also led to an underestimation of the current results [42]. Last, the determinants were measured once at baseline. Patients who had transient deficiency, who developed abnormality over time, or who reverted to normal levels probably due to supplement use during follow-up may have been misclassified.

Conclusions

This study showed that a combination of determinants of anaemia is associated with the presence and development of anaemia in the very old. Since anaemia in the very old is common and associated with poor outcomes [43], health professionals need to be aware that individuals over 85 years of age with a combination of determinants of anaemia present with a high risk of prevalent and incident anaemia. However, further research is needed to explore the underlying pathophysiological mechanisms, either on the specific combinations of determinants, or on the effects within different ethnicities and in younger populations.