Effect of dementia on the incidence, short-term outcomes, and resource utilization of invasive mechanical ventilation in the elderly: a nationwide population-based study

We performed a retrospective population-based study using the Spanish Health Ministry’s National Minimum Basic Data Set (MBDS). This official database gathers information derived from discharge reports from all acute-care hospitals in Spain. For each hospitalization, demographic and clinical information is provided including a main diagnosis, 13 secondary diagnoses, and up to 20 procedures coded at each center before patient discharge according to International Classification of Diseases, Ninth Edition, Clinical Modification (ICD-9-CM) codes as well as corresponding diagnosis-related group (DRG) codes. This information, whose registration is mandatory by law in the National Health System, is considered to be representative of the national population as the database covers over 90% of all annual hospital admissions produced in our country [17].

To calculate incidence rates, we used population data provided by the Spanish Statistics Institute [18]. Hospital admissions data were provided by the Ministry of Health, Resources and Welfare [17]. All data used are anonymous so, according to Spanish law, the need was waived for informed consent [19].

Hospitalizations involving subjects ≥ 65 years who received MV from January 1, 2000, to December 31, 2013, were identified using the ICD-9-CM codes: 96.70 (continuous invasive MV of unspecified duration), 96.71 (continuous invasive MV < 96 consecutive hours) and 96.72 (continuous invasive MV ≥ 96 consecutive hours). These codes are considered specific, stable, and valid [20].

According to ICD-9-CM coding norms, one of these codes is assigned to hospitalized patients who require MV except if used during a routine surgical procedure. Postsurgery MV is coded if lasting more than 2 days or if the clinician reports its duration was longer than planned. MV duration was measured from the moment of endotracheal intubation [17].

Dementia is a clinical syndrome characterized by a global, progressive cognitive impairment that generates functional decline and disability [21]. A variety of brain-damaging conditions, such as degenerative, vascular, metabolic, or toxic brain disease, can lead to dementia, and although the most frequent cause of dementia from middle age to elderly is Alzheimer’s disease, most patients suffer from a mixture of different pathologies [21]. We defined dementia as the presence in any of the diagnoses entered in the database of the ICD-9-CM codes: 290.0–290.9 (dementias), 291.1 (alcohol-induced persisting amnestic disorder), 291.2 (alcohol-induced persisting dementia), 292.82 (drug-induced persisting dementia), 292.83 (drug-induced persisting amnestic disorder), 294.0 (amnestic disorder in conditions classified elsewhere), 294.1 (dementia in conditions classified elsewhere), 294.2 (dementia, unspecified), 294.8 (other persistent mental disorders due to conditions classified elsewhere), 294.9 (unspecified persistent mental disorders due to conditions classified elsewhere), 331.0 (Alzheimer’s disease), 331.1 (frontotemporal dementia), 331.2 (senile degeneration of the brain), 331.7 (cerebral degeneration in diseases classified elsewhere), 331.82 (dementia with Lewy bodies), and 331.9 (cerebral degeneration, unspecified). Similar broad definitions of dementia have been used by others [14] to improve case detection.

To assess the comorbidity burden, we used the Charlson index score validated by Deyo [22] and improved for ICD-9-CM [23] according to secondary diagnoses. For the purposes of this study, dementia itself was excluded in the Charlson Index [24].

For every case, the main diagnostic group was assessed according to the ICD-9-CM chapters: infectious disease (001–139), neoplasms (140–239), endocrine diseases (240–279), hematological diseases (280–289), neurological diseases (320–389), diseases of the circulatory system (390–459), respiratory diseases (460–519), diseases of the digestive system (520–579), diseases of the genitourinary tract (580–629), diseases of the skin and subcutaneous tissue (680–709), diseases of the musculoskeletal system and connective tissue (713–739), and injury-poisoning (800–999).

We conducted a descriptive and comparative analysis of cases with and without dementia, including demographic and clinical information, comorbidity burden, and hospital mortality, stay, and costs. Charlson comorbidity scores are provided as a continuous variable and as categorical with 4 groups (0, 1–2, 3–4, > 4) of increasing severity and impact on outcomes [25]. Categorical variables are expressed as absolute frequencies and percentages while continuous variables are given as geometric means and geometric standard deviation (SD), as geometric means are less influenced by extreme values than arithmetic ones. To test differences in categorical variables, we used Pearson’s chi-square test.

In-hospital mortality was estimated as the number of deaths relative to the number of cases and expressed as a percentage, or case fatality rate (CFR). To examine the effect of dementia on mortality, stay, and costs, we conducted regression models (logistic regressions for mortality and linear regressions of log-transformed stay and costs) and adjusted them in two multivariable models. Model 1 was adjusted for baseline characteristics (age, sex, Charlson index) whereas model 2 also included the principal diagnosis at admission and MV duration. Results are expressed as odds ratios (OR) with 95% confidence intervals (for mortality) and geometric means ratio with 95% confidence intervals (for stays and costs).

We examined temporal trends in MV incidence rate (per 100,000 people and per 10,000 hospital discharges), proportion of MV cases with dementia, prolonged MV (defined as a duration of ≥ 96 consecutive hours according to the ICD-9-CM codification system), and CFR. To this end, we used joinpoint regression models—generalized linear models that assume a Poisson distribution [26]. In these models, any apparent trend can be statistically assessed through a Monte Carlo permutation method [27]. Trends are presented as the average annual percentage change (AAPC), a summary measure of the overall trend over the study period. To compare the AAPC of both study groups, we examined whether their regression mean functions were parallel, allowing for different intercepts, using the Pairwise comparison parallel test. In addition, we analyzed trends in length of stay and costs using Cuzick’s p-trend test.

Rates of MV use referred to hospital discharges and the general population underwent a significantly greater increase among the subjects with dementia compared to those without (Table 3). In patients with dementia, the rate of MV use per 10,000 hospital discharges went up from 1.76 in 2000 to 3.57 in 2013, with an AAPC of 4.7%. Meanwhile, in patients without dementia, the rate of MV use per 10,000 hospital discharges went up from 114.5 in 2000 to 131.3 in 2013 with an AAPC of 0.91%. According to the comparability test, trends of MV in dementia and non-dementia cases were different (p value for test for parallelism = 0.005). The population incidence of MV in people with dementia rose from 3.1 to 6.9 per 100,000 inhabitants ≥ 65 years, yielding an AAPC of 5.39%, while in the non-dementia group it increased from 203.6 to 253.3, for an AAPC of 1.62% (Fig. 1). According to the comparability test, population incidence trends in dementia and non-dementia were also different (p value for test for parallelism = 0.03).

Table 3 also provides temporal trends in the use of VM ≥ 96 h and in-hospital mortality (CFR), indicating that subjects with dementia underwent no significant changes over the study period, while in the non-dementia cohort the use of prolonged MV increased and mortality decreased. As shown in Fig. 2, cases with dementia have not followed the descending trend shown by the cases without dementia; rather, mortality has fluctuated over time.

Figure 3 illustrates that geometric mean hospital stay in the cohort with dementia has steadily declined over time, while it has been relatively stable in the cohort without dementia. Geometric mean hospital costs per case have risen markedly over the study period. This increase has been nevertheless lower among cases with dementia, and inter-cohort differences have persisted (Fig. 3).

Our study has several limitations we should mention. When working with clinical-administrative data, sensitivity to detect the variables of interest depends directly on the discharge report completed by the responsible physician. There is evidence to suggest that dementia has been undercoded in discharge reports, especially in mild or complex cases [39]. To minimize this limitation, we used a broad definition of dementia in line with previous, similar studies [14, 16], even though these definitions have not been validated against clinical charts. Moreover, MV is a major procedure which is easily identified in a patient’s clinical record and whose ICD-9-CM codes are stable and validated [20]. However, we did not have access to staging information of our dementia cases. Given that an inverse relationship has been established between dementia severity and the frequency at which patients are hospitalized and that some clinical guidelines emphasize the need to treat persons with dementia at their homes [37], it is likely that our dementia cohort will contain a high proportion of mild-moderate severity cases. Our data source also prevents us from knowing other individual factors such as pharmacological treatments; the existence of an advance directive or their socio-cultural or educational level which makes it impossible to further characterize this cohort and stratify the results according to these factors. Nonetheless, given the universal character of our national health system with equal access to the whole population, aspects such as the socioeconomic level have not influenced the results obtained. Likewise, given the regulation of our national health system, we can assume that clinical and coding practices have not been related to economic incentives. Also, our database does not include physiology-based scores of common use in ICUs, such as APACHE or SAPS. Notwithstanding, Christensen and colleagues have shown that the Charlson comorbidity index performs similarly to physiology-based scores at predicting short- and long-term mortality for ICU patients [34]. Finally, because of this study’s retrospective nature, we cannot rule out that temporal trends may, at least in part, be associated with different treatment practices during the long period of study. However, the population nature of our study, its main strength, means we can assume a lack of selection bias and can also extrapolate its results. In addition, RECORD recommendations for reporting of results were followed [40].