Prevalence of and Recovery From Anemia Following Hospitalization for Critical Illness Among Adults

Key Points Question Is anemia in patients with critical illness associated with persistent anemia following discharge? Findings In this cohort study of 6901 adults hospitalized for critical illness, 41% had prevalent anemia preceding hospitalization and 74% of those without prehospitalization anemia developed incident anemia, for an overall anemia prevalence of 80% at hospital discharge. Rates of complete recovery from hospital discharge anemia at 12 months posthospitalization for those alive with available hemoglobin assessments were 58% for mild anemia, 39% for moderate anemia, and 24% for severe anemia. Meaning The findings of this study suggest that anemia is common and often persistent in the first year after critical illness.


Introduction
Anemia is a well-known complication of critical illness. Virtually all patients with intensive care unit (ICU) stays longer than 7 days experience anemia, 1 and more than 75% of critical illness survivors are anemic at hospital discharge. 2 Clinicians have generally become more tolerant of anemia given similar short-term mortality rates with restrictive vs liberal red blood cell (RBC) transfusion strategies. [3][4][5] As such, the incidence of transfusion has decreased over time with a corresponding increase in the proportion of patients discharged with anemia. 6,7 Consequences associated with anemia remain incompletely defined, although recent data suggest that anemia at ICU discharge is associated with reduced physical function 3 months later in critical illness survivors. 8 Anemia has also been associated with reduced physical function and quality of life in noncritically ill populations. [9][10][11][12][13][14][15][16] Despite the high prevalence of anemia during critical illness, longitudinal data profiling anemia development and subsequent recovery are lacking. While many critically ill patients are anemic at the time of ICU admission, [17][18][19] it is unclear whether anemia develops abruptly during hospital admission or insidiously before hospitalization, which may have implications for the early identification of high-risk patients before the development of critical illness. Furthermore, data regarding the resolution or persistence of anemia beyond the initial hospital encounter are limited, 20 representing a gap in our understanding of recovery from critical illness.
In this population-based investigation, we examined longitudinal changes in hemoglobin concentrations before, during, and after critical illness and assessed the associations between hospital discharge hemoglobin concentrations and posthospitalization mortality. Moreover, we assessed differences in anemia recovery based on the severity of anemia at hospital discharge, which is information that will be useful in facilitating future efforts assessing the associations between hemoglobin recovery and post-ICU outcomes.

Methods
Residents of Olmsted County, Minnesota, were eligible for inclusion, with medical records obtained through the Rochester Epidemiology Project, a comprehensive epidemiologic database of population health information. 21,22 Inclusion criteria included adult age (Ն18 years), admission to an ICU during hospitalization from January 1, 2010, through December 31, 2016, and hemoglobin concentration available in the first 24 hours of hospitalization (ie, hospital admission hemoglobin concentration, defined as the first hemoglobin concentration measured during admission). For patients surviving critical illness, an independent hospital discharge hemoglobin concentration (ie, the last hemoglobin concentration measured before hospital discharge) must have been obtained in the 5 days before discharge. Survivors missing either of these measurements were excluded, as a principal study objective was to assess changes in hemoglobin concentrations during hospitalization and patients with only a single hemoglobin assessment would be unlikely representative of true critical illness. Only the first hospitalization with an ICU admission was included for each patient, such that no patients were included more than once.
This was a population-based cohort study conducted under institutional review board approvals from the Mayo Clinic and Olmsted Medical Center with waived requirement for written informed consent given minimal patient risk, although consistent with Minnesota statute 144.295 that patients who had previously denied authorization for medical record use in observational research were excluded. The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines were used in study design, conduct, and reporting of results. 23

Outcomes and Assessments
The primary outcomes were changes in hemoglobin concentrations and the prevalence of anemia at 3, 6, and 12 months after hospital discharge. Anemia was defined by hemoglobin values less than 12.0 g/dL for women or less than 13.5 g/dL for men (to convert to grams per liter, multiply by 10), with further classification by severity in accordance with previously used definitions 24,25 : mild (hemoglobin Ն10.0 g/dL and <12.0 g/dL for women or <13.5 g/dL for men), moderate (hemoglobin Ն8.0 and <10.0 g/dL), and severe (hemoglobin <8.0 g/dL). Additional outcomes included the prevalence of baseline anemia preceding hospitalization, identified using the most recent hemoglobin concentration measurement preceding hospital admission (within 12 months); the rate of incident hospital-acquired anemia; changes in hemoglobin concentrations from hospital admission to hospital discharge and ICU admission to ICU discharge; hemoglobin recovery or persistence in the first 12 months after hospital discharge, with complete recovery defined as attainment of nonanemic status; and 12-month posthospitalization all-cause mortality as ascertained through Minnesota Department of Health and National Death Index records.
The primary variables of interest were blood hemoglobin concentrations obtained in the 12 months before hospitalization, during hospitalization with associated critical illness, and in the 12 months after hospitalization. All hemoglobin measurements were extracted using the Rochester Epidemiology Project data repository, which includes all measurements obtained during inpatient, outpatient, and emergency health care encounters in southeastern Minnesota and western Wisconsin (27 counties). 22 Hemoglobin concentration measurements were divided into predefined categories: prehospitalization (ie, in the 12 months before and up to 1 day before hospital admission, with the value closest to hospital admission used to define baseline hemoglobin concentration), hospital admission (ie, in the first 24 hours of hospitalization), ICU admission (ie, in the first 24 hours of ICU admission), ICU discharge (ie, in the 48 hours before ICU discharge; ICU discharge hemoglobin values could not be shared with ICU admission values), hospital discharge (the closest value to hospital discharge, occurring in the 5 days before discharge), and at 3, 6, and 12 months post hospitalization (within 2 months of each interval, without overlap such that any given hemoglobin value could be included only in the closest interval). Mean corpuscular volumes at hospital discharge were used to define anemia as microcytic (<80 μm 3 ), normocytic (80-100 μm 3 ), or macrocytic (>100

Statistical Analysis
Data analysis was conducted from June 1 to December 30, 2019. Patient demographic, laboratory, and admission characteristics were summarized as median (interquartile range [IQR]) for continuous variables and frequency (percent) for categorical variables according to prehospitalization anemia status. Anemia and mortality rates are summarized as number (percent). Given that hemoglobin measurements were not uniformly present in an observational database, not all patients had measurements available at each posthospitalization interval. In a primary approach, changes in hemoglobin concentration and anemia status with corresponding 95% CIs are reported for patients with available data (complete case analysis). Recognizing that missing hemoglobin data may be informative (eg, patient doing well without need for recheck, censored owing to death or change of residence, and transitioned to hospice care), we also provide comparative demographic and clinical characteristics by the number of posthospitalization hemoglobin concentration assessments (centered above and below the median value of 5: 0, 1-5, and >5). In addition, multiple imputation to account for missing values was conducted using 50 independent data sets and fully conditional specification, including age, sex, admission APACHE III score, admission type, hospital length of stay, log(follow-up time), last follow-up status (alive/dead), RBC transfusion during hospitalization, non-RBC transfusion during hospitalization, and available hemoglobin concentration values at hospital admission, ICU admission, hospital discharge, and 1, 3, 6, 9, and 12 months post hospitalization, with the values then converted to corresponding anemia status for those alive and not censored owing to change of residence. Assessments of anemia development and recovery are presented for the full cohort and subgrouped for surgical and nonsurgical patients (ie, no surgery during hospitalization). Multivariable Cox proportional hazards regression models were used to assess the associations between hospital discharge hemoglobin concentrations and instantaneous hazard of death in the first year post hospitalization. Potential confounding variables selected a priori and included in the model were age, sex, admission type, Charlson Comorbidity Index score, admission APACHE III score, ICU length of stay, and prehospitalization anemia status (none, mild, moderate, severe, and not available). Schoenfeld residuals were inspected visually to assess the assumption of proportional hazards. Hazard ratios (HRs), 95% CIs, and associated P values per 1-g/dL increase in hemoglobin value were reported. P values <.05 (2-sided) were considered statistically significant. All analyses were performed using SAS, version 9.4 (SAS Institute Inc). Prehospitalization hemoglobin concentrations were available in 5694 patients (83%), with median time from measurement to hospitalization of 1.2 (IQR, 0.6-3.6) months. The median prehospitalization hemoglobin value was 13.1 (IQR, 11.6-14.4) g/dL; 2320 patients (41%) with available data had prevalent anemia before hospitalization. Those without prehospitalization hemoglobin concentration data and nonanemic patients were generally healthier than those with anemia, as evidenced by younger age and lower Charlson Comorbidity Index, SOFA, and APACHE III scores (

Anemia During Hospitalization
The median change in hemoglobin from prehospitalization to hospital admission was −0.7 (IQR, −1.8 to 0.2) g/dL ( Table 2). Of those who were previously nonanemic (n = 3374), 40% developed incident anemia within the first 24 hours. In total, 56% of patients were anemic within 24 hours of ICU admission, with this proportion increasing to 78% by ICU discharge, exclusive of 3% dying during ICU admission. The rate of incident anemia at ICU admission was 44% and the rate at ICU discharge was in Figure 1 and eTable 1 in the Supplement and by type of admission (surgical vs nonsurgical) in eFigure 2 in the Supplement. Of those who were nonanemic before hospitalization and survived to hospital discharge (n = 3204), only 26% remained nonanemic (74% incident anemia; 66% mild, 32% moderate, and 2% severe).

Discussion
In this large, population-based investigation of critical illness survivors, 41% had anemia before hospital admission, with this number increasing to 80% at hospital discharge. The rate of incident anemia during hospitalization was 74% despite relatively short ICU (median, 1.2 days) and hospital (median, 4.8 days) lengths of stay. At 12 months post hospitalization, nearly half of critical illness survivors with available hemoglobin concentration data remained anemic, including more than one-quarter of survivors without prehospitalization anemia. Higher hospital discharge hemoglobin concentrations were associated with reduced postdischarge mortality.
Beyond this work, there is a paucity of data regarding recovery from anemia in survivors of critical illness. In a prospective investigation of 19 survivors with anemia at ICU discharge, more than 50% experienced persistent anemia at 6 months. 20 This anemia was accompanied by elevated concentrations of inflammatory markers, including interleukin-6 and C-reactive protein, suggesting the presence of a persistent inflammatory state that may impair erythropoiesis. This inflammatory state has been proposed as a possible factor associated with delayed recovery of physical function after critical illness, 26 and anemia has also been associated with reduced physical function (ambulatory capacity and activities of daily living) in critical illness survivors. 8 In noncritically ill older persons, anemia has been associated with a variety of impairments in daily functioning, including decreased mobility, 11,27,28 muscle weakness, 11,12 falls, 29 impaired physical rehabilitation, 30 reductions in patient-reported physical function, 12 and global cognitive decline. 31 With a growing awareness of functional limitations in critical illness survivors, 32 the potential associations between persistent anemia and functional recovery in this high-risk group should be considered.

Strengths and Limitations
Strengths of this investigation include a population-based approach with large sample size and longitudinal ascertainment of hemoglobin concentrations before and after critical illness. As such, this work provides a framework for future studies addressing anemia in the critically ill population, including assessment of the associations between hemoglobin recovery and clinical recovery, identification of risk factors for persistent posthospitalization anemia, and evaluations of the safety and efficacy of anemia management interventions on posthospitalization clinical outcomes. Previous investigations not limited to the critically ill have noted associations between hospital-acquired anemia and in-hospital mortality 33 and 30-day hospital readmission rates. 34 Building on this information, we found an inverse association between hospital discharge hemoglobin concentrations and mortality after hospitalization, suggesting that anemia attenuation strategies may have important consequences. The prevalence of severe anemia at hospital discharge increased throughout the study period, suggesting increasing tolerance for anemia in clinical practice.
There are limitations of this study. First, we used historical data, and anemia-related diagnostic studies were unavailable for many patients; hence, the causes of anemia were not ascertained. Only 5% of patients received pharmacologic anemia treatment (ie, iron or erythropoiesis-stimulating agents). Second, hemoglobin values were not uniformly available post hospitalization, which may represent random or nonrandom (ie, informative) missingness such that missing data could be indicative of healthier patients not requiring health care contact, those lost to follow-up, or those with significant underlying illness who have died or transitioned to less-aggressive medical care settings. This lack of data impacts our ability to fully assess patterns of anemia recovery, although several approaches were taken to address this issue. Patients with change of residence or death were censored, although it is probable that those who died (14% of hospital survivors) would be less likely to recover from anemia. In addition, clinical characteristics were compared between patients with and without posthospitalization hemoglobin assessments. Those without hemoglobin assessments were generally similar to those with 1 to 5 assessments but were younger and had less severe disease than those with more than 5 assessments, which could lead to underestimation of anemia recovery.
Multiple imputation including observed covariates and available hemoglobin assessments was used to impute missing posthospitalization hemoglobin data, with results that were generally consistent with the complete case analysis (45% observed vs 41% estimated anemia prevalence at 12 months).
As a third limitation, generalizability to other critically ill populations is unclear. For example, ICU and hospital durations were relatively short in this investigation, which is likely reflective of a propensity to provide intensive care for patients who may be deemed to be lower acuity in other health care and geographic locations. It is possible that anemia recovery may be further impaired in those with longer hospitalizations. In addition, these results were derived from a population that is predominantly of European descent with reliable access to health care, which again may limit generalizability.

Conclusions
Incident anemia occurs in nearly three-quarters of patients hospitalized for critical illness and often persists through the first year post hospitalization. Further studies are necessary to more fully evaluate associations between posthospitalization anemia and patient-important clinical outcomes.
In addition, studies are warranted to distinguish patients likely to recover from anemia after hospitalization from those who may experience prolonged anemia; the latter group may benefit from close outpatient follow-up and/or targeted anemia management strategies.