A nationwide study of the long-term prevalence of dementia and its risk factors in the Swedish intensive care cohort

Dementia is a common and often detrimental group of diseases with a sharply increasing incidence in the elderly [1, 2]. In general, the disease is characterized by memory loss, disturbances in language, altered perception, and other psychological and psychiatric changes. Together, these symptoms cause impaired daily functioning [3] and may severely affect health-related quality of life [4].

In hospitalized patients, and especially in the intensive care population, sepsis is a common syndrome [5] defined by a dysregulated host response to an infection [6]. The systemic inflammation in sepsis has been suggested to have a long-term negative impact on the brain [7]. Both short- and long-term effects of experimental sepsis on brain cells and behavior have been described in rodents [8‐10]. Septic encephalopathy and persisting cognitive disturbances have also been reported in human studies [11‐13]. Moreover, in patients with an assessment of cognitive function before and after hospitalization, a decline in cognitive function was more pronounced after hospitalization for severe sepsis than after admission for other reasons [14]. Sepsis diagnosis has also been reported to be more common in the history of patients with dementia than age- and sex-matched controls from the health care system [15]. Finally, sepsis is an independent risk factor for dementia in observational studies [16, 17]. Based on these data, we hypothesized that dementia would develop more commonly in patients admitted for, or developing, sepsis in intensive care compared with other patient groups.

However, the prevalence of dementia in the population is relatively low [1]. Additionally, dementia is usually a slowly developing syndrome with a long subclinical period before diagnosis [18]. Dementia also increases the risk of sepsis [19], and several comorbidities are risk factors for both dementia and sepsis.

We therefore set out to investigate our hypothesis in a large nationwide database with an extended follow-up and accounting for comorbidities.

Our primary endpoint was the hazard ratio (HR) of sepsis for a diagnosis of dementia adjusting for several potential risk factors. We also investigated the impact of these risk factors and crude dementia incidence in this cohort.

This study was approved by the Regional Ethics Committee of Uppsala (approval no. 2016/421). Since this is a registry-based study, informed consent was waived. The protocol of the study was registered a priori with the Australian and New Zeeland Clinical Trials Registry (registration no. ACTRN12618000533291). Reporting strictly follows the STROBE Statement [20].

Of 315,155 patients, 210,334 (67%) were still alive and without dementia 1 year after ICU admission (Fig. 1). Of those 210,334 patients, 16,115 (8%) had a sepsis diagnosis code in ICU care. The patients were followed for a median of 3.9 years (IQR 1.7–6.6). SAPS3 data were completely missing in 23.8% of the patients in the Sepsis cohort and 45.2% in the No sepsis cohort. Of patients with missing SAPS3, 46% had a registration of an APACHE II score in the Sepsis cohort and 23% in the No sepsis cohort. Missing SAPS3 data were imputed. Of all patients admitted to the ICU in 2005–2016, 8495 (4.0%) emigrated at least once from Sweden, and of these, 5358 (63.1%) had at least one listing in the NPR or the SveDem > 1 year after ICU admission.

Patients in the Sepsis cohort were older, had higher SAPS3, and had more comorbidities, expressed as a higher CCI score. They also had longer ICU-LoS and H-LoS and were more commonly treated with invasive ventilation and RRT (Table 1). During follow-up, the Sepsis cohort had a mean of 3.9 episodes of inpatient care, whereas 32% of these patients had no such episode. The No sepsis cohort had a mean of 3.4 episodes of inpatient care, and 38% of these patients were without any such episode.

The 6312 patients ultimately developing dementia were older and predominantly female and had higher SAPS3 and longer ICU-LoS and H-LoS than those without dementia (Table 2). In addition, those patients who developed dementia were less frequently treated with RRT and invasive ventilator therapy. Finally, patients with dementia had a higher rate of acute surgical admissions, but a lower rate of planned surgical admissions. The size of these differences was generally small, however.

Dementia prevalence and 1-year mortality increased with age in patients admitted to the ICU during the study and alive on the last day of follow-up (December 31, 2016) (Fig. 2). In the unadjusted analysis, dementia was more common in individuals diagnosed with sepsis during their ICU stay (log-rank p < 0.001) as depicted in the Kaplan-Meier survival curve (Fig. 3). However, after adjusting for age, sex, CCI score, SAPS3 box 2+3, H-LoS, ICU-LoS, invasive ventilator therapy, and RRT, sepsis was no longer an independent risk factor for dementia (HR 1.01, 95% CI 0.91–1.11) (Fig. 4). Only age, SAPS3 box 2+3, RRT, and ventilator therapy were independent risk factors in the model, with the HR for dementia increasing with increasing age and SAPS3 box 2+3 but decreasing with the use of RRT and ventilator therapy.

In our nationwide Swedish cohort of 210,334 patients alive without dementia 1 year after ICU admission, sepsis was found to be a crude risk factor for a later diagnosis of dementia. However, after adjusting for baseline characteristics of the patients in our cohort, sepsis was no longer an independent risk factor for dementia. This finding was consistent in all performed sensitivity analyses.

In a previous study [36], 516 patients having survived a sepsis episode were compared with 4517 patients having survived a hospitalization without sepsis. All included individuals underwent at least one prospective cognitive testing. When followed for up to 1 year, the sepsis patients performed worse on repeated cognitive testing. However, it is not clear whether this condition evolves into fulfilling the diagnostic criteria of dementia, and the authors did not control for age, comorbidities, or the degree of acute illness. In the present study, we controlled for several factors expressing both comorbidity and the degree of acute illness besides age. Guerra et al. performed two studies on a 2005 Medicare cohort. In their first study [16], the findings seem to confirm the hypothesis that the higher rate of observed cognitive dysfunction after sepsis is evolving into a higher rate of clinically diagnosed dementia in patients treated in the ICU with sepsis than in those treated in the ICU without sepsis during their hospitalization. However, they were only able to adjust for comorbidities diagnosed in the year preceding the index hospitalization, thereby risking underestimating the comorbidities. In addition, the authors run the risk of overlooking the presence of dementia diagnoses in earlier years that was not registered in the year preceding hospital admission. Furthermore, because dementia is a syndrome of a slowly evolving disease [18], dementia diagnosed early after the index hospitalization might be an example of a clinically overt disease coming to the attention of the medical system in the convalescence period after hospitalization rather than a consequence of the sepsis episode or acute illness per se. Dementia may also be a risk factor for sepsis [19]. We sought to lessen the effect of both over-diagnosis due to hospitalization and causality problems between sepsis and dementia by excluding dementia diagnoses registered during the first year after ICU admission. In the second study by Guerra et al. [17], patients admitted to the ICU with sepsis were compared with non-hospitalized patients matched on age group, sex, and race. Sepsis was a significant risk factor for dementia. However, in a model adjusting for comorbid diagnoses associated with dementia during the index hospitalization, the effect of sepsis decreased compared with using the same comorbidities diagnosed before the index hospitalization. Moreover, when using comorbidities diagnosed during the index quarter in the model, the effect of sepsis on the risk of dementia disappeared in line with our findings. In a case-control study [15], the odds ratio for having had a sepsis diagnosis in 5955 patients with a dementia diagnosis was higher than in age- and sex-matched controls without dementia. In their design, the authors did not account for the amount of time elapsed from sepsis to the diagnosis of dementia, nor did they adjust for comorbidities diagnosed during the index hospitalization.

Despite that the patients in the Sepsis and the No sepsis cohorts are from the same ICU cohort, they were not comparable, i.e., sepsis cohort patients had more chronic illnesses, were older, and had a higher degree of acute illness. Hence, it was essential to adjust to these specific factors. Our study used the revised CCI as a composite measure of the comorbid burden of each patient. We also adjusted for the severity of acute illness, as we presumed that it might mediate the effect of sepsis on dementia development. SAPS3 box 2+3, i.e., acute illness severity, was an independent risk factor for dementia. However, in a sensitivity analysis in which SAPS3 box 2+3 was excluded from the model, sepsis was not a significant risk factor for dementia. This finding implies that SAPS3 box 2+3 does not modulate the effect of sepsis in the model.

As expected, age was a strong risk factor for developing dementia during follow-up in our cohort of ICU-treated patients. Intriguingly, in the oldest age category (> 90 years), we observed a lower prevalence of dementia compared with another study on a Swedish cohort [33]. This lower prevalence may be related, in part, to the very high mortality rates in this elderly patient group that have been treated in the ICU. Surprisingly, the CCI score was not an independent risk factor for dementia in the model. We speculate that this finding is due to the small difference between the 25th and 75th percentiles of the CCI score. A wider range of CCI might have yielded another result. For the variables in which cubic splines were applied, the HR was calculated for the difference between values at the 25th and 75th percentiles to reduce the risk of false conclusions from the HRs of these splined variables. Still, despite this adjustment, results for the splined variables need to be interpreted cautiously. However, their validity as covariates of sepsis—the primary endpoint of the study—in the model is higher after cubic spline application than if we would have categorized these nonlinear variables [37].

In conclusion, although dementia was more common in the whole nation Swedish ICU cohort for 2005–2015 treatment for sepsis in the ICU, sepsis was not a risk factor for later dementia after adjustment for pre-specified, relevant, baseline variables. In our sample, acute illness severity altered the risk of dementia, which might account for a fraction of the apparent causality between sepsis and dementia in ICU patients.