The relationship between ICU survivorship, comorbidity and educational level in quality of life after intensive care

ICU survivors have lower quality of life (QoL) compared to a non‐ICU‐treated population. The reason for this is not fully understood, but differences in baseline characteristics may be an important factor. This study evaluates the roles of comorbidity and educational level as possible explanatory factors for differences in QoL in ICU survivors compared to a non‐ICU‐treated population.

an association from belonging to the ICU survivor group.Comparing QoL in ICU survivors to that of a non-ICU-treated population may be adequate despite differences in baseline characteristics.
K E Y W O R D S critical care, intensive care unit, critical illness, quality of life, follow-up studies, long-term adverse effects, survivors, survivorship, socioeconomic status

Editorial Comment
Many factors can contribute to reported quality of life among ICU survivors.This questionnairebased study assessed possible contributions of comorbidity and level of education in this context.These findings demonstrate the complexity in identifying factors which may contribute to reported quality of life in the post-ICU cohort.

| INTRODUCTION
Impaired quality of life (QoL) after intensive care has been reported in several observational studies. 1,2Survivors suffer from reduced physical 3 and mental health, 4 cognitive dysfunction, 5 and difficulties returning to work. 6An association with critical illness and complications from advanced treatment has been postulated in the post-intensive care syndrome (PICS). 75][16][17] Consequently, comparisons between ICU survivors and non-ICU-treated populations could be misleading if these differences are not managed properly.
In this study, we explored ways in which ICU survivorship, comorbidity and educational level relate to QoL, specifically focusing on how these factors interact with each other.For this purpose, we used our previously described comparison between an ICU survivor group and a non-ICU-treated control group where long-term QoL was measured with a provisional questionnaire, and where the findings in this study will aid a future reduction of the number of questions in the finalisation of the questionnaire. 18This study aims at assessing the roles of comorbidity and educational level as possible explanatory factors for differences in QoL in ICU survivors compared to a non-ICU-treated population, with a specific focus on potential effect modification by these variables.

| Study setting and populations
In a first step, 32 adult ICU survivors were interviewed at a minimum of 6 months after ICU discharge.In total, 195 unique issues were extracted from the interviews and rephrased as questions.All questions were subsequently organised into 13 domains and compiled into a questionnaire.In a second step, this questionnaire was used in a cross-sectional study comparing responses between long-term ICU survivors and a non-ICU-treated control group.A full description of both steps has been published elsewhere. 18Eligible participants were all adult ICU survivors admitted between February 2013 and December 2015 to one of three mixed ICUs at a university hospital in Gothenburg, Sweden, with a minimum ICU length of stay of 72 h.
Patients with neurological/-surgical admission diagnosis were excluded.Follow-up was performed 6 months to 3 years after ICU discharge.A non-ICU-treated control group was randomly selected from the Swedish Population Registry, matched by age and sex.An additional question regarding previous ICU care was added to the questionnaire for the control group and constituted an exclusion criterion.
Reporting adheres to the STROBE guidelines and recommendations within intensive care literature (Table S1). 19,20

| Definitions
The comorbidity variable was defined as the number of comorbidities (0 to 19; Table 5 in reference 18) when filling in the questionnaire.
Educational level was used as a proxy variable for socioeconomic status (SES) and constructed as a five-level scale from 0 to 5 (None -Elementary school -Vocational school -Upper secondary school/High school -University/College) as answered when filling in the questionnaire.
We analysed the following associations simultaneously on a question-per-question basis (Figure 1): The association between group belonging (ICU survivor group or non-ICU-treated control group) and QoL (arrow a).The association between either comorbidity or educational level and QoL (arrow b).A potential moderating effect by either comorbidity or educational level on the association between group belonging and QoL (arrow c).A moderating effect implies that the effect on QoL from belonging to the ICU survivor group varies with the degree of comorbidity or educational level, that is, effect modification.This model allows for multiple associations to exist in parallel instead of being mutually exclusive.Finally, the model will identify distorted association (collider bias) between group belonging and QoL (arrows d). 21

| Statistical analysis
A total of comorbidities was calculated for all respondents by an independent samples T-test with 5000 bootstrap iterations.Independent bivariable correlation analyses with one analysis per question comparing responses from the ICU survivors and the non-ICU-treated control group were initially conducted.3][24] Results from the multiple linear regression analysis are presented as unstandardized coefficients whereas multiple binary logistic regression results are described in terms of the log odds.In a first round, we conducted a series of regression analyses with a dichotomous predictor variable (ICU survivor group and non-ICU-treated control group), with the response to the QoL questions as the outcome variable and comorbidity as a moderating variable.In a second round, educational level replaced comorbidity as the moderating variable.In each analysis, comorbidity and educational level were treated as moderating, continuous, variables (i.e., not independent categorical variables).The product terms in the moderation models were mean-centred and bootstrapped (n = 5000) to reduce bias and to address non-normal distributions.The multivariable analyses provide estimates of the separate effects of the independent variables (ICU survivor group/non-ICU-treated control group and comorbidity; ICU survivor group/non-ICU-treated control group and educational level) as well as of a potential moderating effect.No correction for multiplicity was needed, considering both the purpose of explaining the associations in one separate hypothesis per item, thus with each question/analysis interpreted independently, and that moderating variables were treated as continuous variables. 25,26P values <.05 were considered statistically significant.No correction for missing data was performed, considering that approximately 5% of data were missing. 27Statistical analyses were conducted with SPSS 27.0 (IBM, Armonk, NY).PROCESS version 3.5 add-on for SPSS was used to explore potential moderating effects.

| Study populations, demographics, and characteristics
Of 518 included ICU survivors and 289 included controls, 395 ICU survivors and 197 controls returned a questionnaire (response rate 76.2% and 85.3%, respectively; CONSORT diagram can be found in Of the comorbidites asked about, hypertension, angina pectoris, myocardial infarction, heart failure, lung disease, pulmonary embolus, mental disease, diabetes, kidney disease, bowel disease, need for walking aid, and amputated limbs were significantly more common in the ICU-survivor group.A range from 0 to 9 comorbidities was present for the ICU survivor group.Among them, 107 survivors (27%) had a single comorbidity, followed by 89 (23%) with two comorbidities.No comorbidities were present in 84 ICU survivors (21%), while only two (1%) had nine comorbidities.In the control group, 84 individuals (45%) reported no comorbidities, 63 (32%) had one, and 26 (13%) had two comorbidities.The average number of comorbidities for the ICU survivor group was 1.94 (95% CI: 1.77-2.12)compared to 0.97 (95% CI: 0.82-1.16)for the control group.There was no significant difference in educational levels between the two groups.

| Bivariable analysis
The initial analysis identified a significant difference in QoL between ICU survivors and non-ICU-treated controls in the responses to 170 of 218 (78%) questions.Responses to all questions in the domains of physical health, gastrointestinal functions and work life differed between the two groups, as did most in pain (95%), ADL (94%), fatigue (86%) and mental health (86%).Urinary tract functions and appetite   2G, Table S2a; Figure 3G, Table S2b).Questions with non-significant differences in responses in the initial bivariable analysis will not be further discussed.

| Impact of comorbidity on quality of life
After introducing comorbidity into the multivariable model, the following could be seen: associations between belonging to the ICU survivor group and QoL remained in 139 of the 218 questions (Figure 2A, B).Of these 139 questions, 80 lacked an association between comorbidity and QoL (Figure 2A), while there was an association between comorbidity and QoL in parallel to that between  2B).
Comorbidity moderated the association between group belonging and QoL in six questions (Figure 2c,d); two questions in cognition and one each in the domains of fatigue, ADL, sexual health, and work life (Table S2a).In 25 questions, there were no longer any significant differences between belonging to the ICU survivor group and the non-ICU-treated control group (Figure 2E,F) after introducing comorbidity as a moderator.
The domains in which the proportionally highest number of questions remained significant after introducing comorbidity as a moderator were appetite and alcohol (6 of 6 questions still with significant differences between groups), physical health (29 of 30 questions) and fatigue (11 of 12 questions).Conversely, the domains with the proportionally lowest number of questions remaining significant after introducing comorbidity into the model were cognition (9 of 18 questions still significant) and urinary functions (2 of 4 questions).

EDU
Regarding a separate effect of comorbidity, parallel to the effect of belonging to the ICU survivor group, large differences could be seen across domains: Sexual health and Sensory functions showed a parallel association between comorbidity and QoL in a majority of issues; 5 of 6 questions and 10 of 13 questions, respectively.Conversely, none of the questions in the domains of cognition and fatigue had a significant and parallel association between comorbidity and QoL.Work life, psychological health, and GI functions showed a similar pattern, with only 1 of 8 questions, 2 of 15 questions, and 1 of 5 questions, respectively, having an association between comorbidity and QoL in parallel to the association between group belonging and QoL.

| Impact of educational level on quality of life
The model using educational level as a moderator showed the following: The significance of belonging to the ICU survivor group remained in 139 questions (Figure 3A, B).Educational level was not associated with QoL in 113 of these 139 questions (Figure 3A).In 26 of the 139 questions, educational level was associated to QoL in parallel to an association between belonging to the ICU survivor group and QoL (Figure 3B).Educational level moderated the association between belonging to the ICU survivor group and QoL in 34 questions, of which 31 had been significant also in their respective bivariable analysis (Figure 3c,d); 11 questions in the domains of Physical health, seven in ADL and occasional questions in all other domains apart from cognition and appetite and alcohol, in which no moderation effect was found (Table S2b).
The domains in which the proportionally highest number of questions remained significant after introducing educational level into the model were cognition (18 of 18 questions still with a significant difference between groups), appetite and alcohol (6 of 6 questions), pain (17 of 18 questions), sensory functions (14 of 15 questions) and fatigue (11 of 12 questions).Conversely, the domains with the proportionally lowest number of questions remaining significant were Urinary functions (2 of 4 questions still significant), ADL (8 of 15 questions) and Physical health (18 of 30 questions), all in which the changes were mainly due to moderation effects (Table S2b).
Overall, associations between comorbidity and QoL were more common than between educational level and QoL.

| DISCUSSION
In this cross-sectional study comparing QoL in ICU survivors 6 months to 3 years after ICU discharge with a non-ICU-treated control group, matched for age and sex and randomly selected from a national population registry, we explored how comorbidity and educational level affected QoL.The main finding was that the higher comorbidity burden in the ICU survivor group compared with the control group did not explain differences in QoL.Similarly, except in two clusters of items within the domains of physical health and ADL, educational level rarely explained differences in QoL between the two groups.A second important finding was that in issues where comorbidity or educational level affected QoL, they mainly did so in parallel to the effect on QoL from belonging to the ICU survivor group.
Our findings differ from studies claiming comorbidity to be the main explanation for low QoL after intensive care.In a prospective, longitudinal study on 980 adult ICU survivors, those with pre-existing disease had a lower QoL score on EQ-5D and SF-36 compared to both previously healthy ICU survivors and a reference group with or without comorbidity. 10 and MCS scores at both measurement points. 28In line with our findings, an analysis on more than 2000 patients from two international randomised multicentre trials in sepsis patients concluded that poor QoL was not predicted by the presence of chronic disease per se but rather due to critical illness. 29These findings persisted in a sensitivity analysis on patients without pre-existing chronic disease when adjusted for the potential confounding effect of comorbidity.
In our study, educational level rarely explained differences in QoL apart from a number of issues within the domains of physical health and ADL.This is in line with a recent study showing a correlation between educational level and in particular the ADL component of PICS in impaired health-related QoL after intensive care. 30The structure and reasons for these findings need to be further studied.Our results within the domain of Cognition differ from studies correlating a low educational level with cognitive impairment after intensive care.
The lack of association between level of education and cognitive dysfunction in our study differs from two recent multicentre studies from the same group in which a higher level of education was protective against cognitive impairment, measured with a battery of tests at 3 and 12 months after hospital discharge, and associated with less PICS problems. 31,32However, neither of these studies had a control group, and as in our study, there was no data on pre-ICU status.

| LIMITATIONS
Our study has limitations.This is a secondary analysis, with only certain variables available.Had the primary goal of the original study been to explore the role of socioeconomic status, additional factors apart from educational level would have been registered. 33For example, in an analysis of two large longitudinal studies on mobility, ADL and psychological distress, various indicators of socioeconomic status were tested, and although education was associated with late-life health, income was a stronger indicator. 34We decided not to ask for income in our provisional questionnaire since it could be a sensitive question affecting response rates.Further, with a mean age of 65 years, too few participants had ongoing employment, another valid marker of SES.We only had data on the number of comorbidities, not the severity.While some studies have found the number of comorbidities directly associated with morbidity in respiratory disease, others have found such an association specifically weak in elderly patients. 35,36Although the study ruled out some factors as explanatory for the differences in QoL, the ICU survivor cohort may have had unidentified reasons for a lower QoL before being admitted to hospital that could have been aggravated by critical illness.Regarding generalizability of the findings, it should be noted that only survivors with a minimum of 72 h ICU length of stay were included, thus caution is required before interpreting these findings to be applicable to all ICU survivors.Furthermore, with regards to the response rate of 76% in the ICU survivor group, we do not have data on how non-participants differ from participants.Finally, median follow-up time for the ICU survivors was more than a year after ICU discharge.However, the timeframe in most questions was "the past month" to compensate for possible response shift.

| Strengths
By using comorbidity and educational level as moderators, we have separated the potential impact of these factors on QoL from that of being an ICU survivor, showing that they can exist in parallel.This methodology is similar to that in a recent database study on mental health before and after intensive care, where intensive care was associated with an increase in mental disease compared to nonhospitalised controls and hospitalised, non-ICU-treated patients. 37th studies show that a comparison with a control group from a non-ICU-treated population may be valid even if there are significant differences in comorbidity.

| CONCLUSIONS
Differences in comorbidity between ICU survivors and non-ICU-treated controls do not explain differences in QoL.Similarly, educational level only rarely explains differences in QoL between these two groups.Instead, in most issues where comorbidity or educational level affects QoL, this effect is parallel to the effect of being an ICU survivor.Relevant comparisons of QoL can be made between ICU survivors and control groups from non-ICU-treated populations if the possibility of parallel effects is taken into account.

F
I G U R E 1 Schematic model of analysed relationships.All analyses are performed on a question-per-question basis.X: Group belonging (ICU survivor group or non-ICU-treated control group); Y: QoL issue (question in questionnaire); Z: Comorbidity variable or Educational level variable; a: an association between being in the ICU survivor group compared to being in the control group and a QoL issue; b: an association between the comorbidity or educational level variable and a QoL issue; c: a moderating effect from the comorbidity or educational level variable on the association between group belonging and the QoL issue; d: collider bias.and alcohol were the domains with the least number of differing responses between the ICU survivor group and the non-ICU-treated control group, 50% and 55%, respectively.All questions are shown in

Table S2a (
comorbidity) and TableS2b(educational level).Seven questions in the comorbidity model and four in the educational level model were subject to collider bias (Figure Belonging to the ICU survivor group compared to the control group is associated with QoL.• Comorbidity has no effect on QoL.QuesƟons with an associaƟon between ICU survivorship and QoL Comorbidity funcƟons as a moderator, i.e. the difference between ICU survivors and controls in QoL varies with the degrees of comorbidity.• Comorbidity is associated with QoL.
EducaƟonal level funcƟons as a moderator, i.e. the difference between ICU survivors and controls in QoL varies with the degrees of educaƟonal level .•EducaƟonal level is associated with QoL .