Frailty index and its association with the onset of postoperative delirium in older adults undergoing elective surgery

Briefly, the PAWEL study was conducted in five medical centers in the southwest of Germany between July 2017 and April 2019 with the aim to evaluate an individualized, multi-professional and multimodal delirium and postoperative cognitive dysfunction (POCD) prevention program. Study setting and design have been described in detail in a previous paper [15]. Patients ≥ 70 years undergoing a major elective surgery with a cut-to-suture time ≥ 60 min were eligible, with the exception of those requiring emergency medicine, with an expected survival ≤ 15 months, or who (or the legal guardian) did not provide written consent. A total of 1470 patients were recruited. For this analysis only data from the control group, defined as PAWEL-R (n = 899 with no delirium intervention), was considered to prevent potential bias due to intervention. Due to missing data occurring in 1.8% of the 30 variables considered for the FI, 198 participants had to be excluded (Additional Table 1). In total, 701 participants could be considered for further analysis. The majority of this group was recruited in orthopedic/other surgical wards (n = 405, 57.8%), while 296 (42.2%) patients underwent cardiovascular surgery.

Ethical approval for the PAWEL Study was obtained from the Ethics Commission of the Faculty of Medicine of the Eberhard-Karls University and University Hospital Tuebingen (517/2017BO1), Ulm University (425/17) and the Ethics Commission of the University of Potsdam (38/2017). All participants gave written informed consent. PAWEL and PAWEL-R (sub-) study were registered on the German Clinical Trials Register (number DRKS00013311 and DRKS00012797, first regristration November 23^rd 2017).

Preoperatively assessments were performed with interview-based self-reports by trained assessors containing questions about current living situation, falls, mobility, subjective memory impairment and weight loss. Age, sex, smoking state, alcohol consumption, education and type of surgery were defined as co-variables. Smoking was categorized into non-smoker, ex-smoker and current smoker. An alcohol-score was built based on the self-reported consumption of beer, wine and liquors (options between never = 0 and daily = 5) using the following formula: alcohol score = 0.05 × beer consumption + 0.12 × wine consumption + 0.4 × liquor consumption taking into account the percentage of alcohol of a standard drink [17]. Education was categorized into ≤ 10 or > 10 years of education. Comorbidities were evaluated as follows: diabetes mellitus, arterial hypertension, cardiovascular disease (myocardial infarction, heart failure, circulatory disorders in the lower extremities, arrhythmia, stroke and/or cerebral hemorrhage or hemiplegia), chronic lung disease, dementia, osteoarthritis, urinary incontinence, tumor (lymphoma, leukemia, metastasized solid tumor or any tumor disease), neurological diseases (M. Parkinson and/or seizures) and liver disease. For the emotional and mental state, the seven questions of the SF-12 were considered: General health, moderate activities, climbing several flights of stairs, interference of pain with daily activites, having a lot of energy, feeling downhearted and blue and interference of physical health or emotional problems with social activities [18]. Visual and hearing impairment were evaluated with a visual acuity test and a whisper test, respectively. Body mass index (BMI) was calculated using self-reported weight and height. The preoperative creatinine levels were implemented in the CDK-EPI formula [19] to estimate the glomerular filtration rate (GFR). For hemoglobin World Health Organisation recommended and sex-stratified cut-offs for anemia were used [20]. Polypharmacy was present if five or more medications were taken [21]. ADLs were assessed using the Barthel Index [22]. For secondary analysis timed-up-and-go (TUG) and handgrip strength were considered. TUG reflected participant`s time needed to rise from a chair, walk 3 m and return into a sitting position [23]. Assistive devices (e.g. walking stick) were allowed. Cut-offs for TUG were defined as follows: 0 pt.: < 10 s., 0.25 pt.: ≥ 10 and < 20 s., 0.75 pt.: ≥ 20 and < 30 s., 1 pt.: ≥ 30 s. [23]. Handgrip strength was measured with a JAMAR hand dynamometer testing both hands. The second measurement on the stronger one was considered in the FI. Differentiating by sex a deficit was assumed by < 16 kg in women and < 27 kg in men [24].

The incidence of POD was assessed at bedside by performing a daily Confusion Assessment Method (I-CAM) [25, 26] during the first postoperative 7 days. In addition a chart review at discharge was performed by trained research physicians applying the DSM-5 delirium criteria as reference standard [27, 28]. POD was a combined endpoint allocated by a positive I-CAM on any postoperative day and/or detection of delirium during chart review [15].

For descriptive analyses categorical variables were expressed as numbers and percentage, continuous variables as mean (standard deviation) or median (Q1, Q3). We proceeded to identify the top ten contributors to the FI among those undergoing cardiac and orthopaedic/other surgical procedures respectively. We used logistic regression models to evaluate the association between frailty and the onset of delirium using the FI as a continuous variable as well as a categorical one. Subjects with a FI ≥ 0.2 were defined as frail [29]. Model 1 adjusted for sex and age; Model 2 adjusted for sex, age, alcohol consumption, smoking, education and type of surgery. We examined the presence of effect modification by sex, and by type of surgery without significant findings (all p-values > 0.2). Odds ratios (OR) with their 95% confidence interval (CI) as well as the calculated c-statistic are reported. A secondary analysis excluding those with a length of stay < 7 days, and missing data on postoperative delirium at 2-months follow-up (n = 672) was performed. In a subsample of 517 participants without missing data for functional measurements we built a 32-items FI enriched with the TUG and handgrip strength. In order to evaluate the impact of adding functional measurements on the predictive value of the FI a secondary analysis was performed contrasting the results of the 30-items versus the 32-items FI. Additional Fig. 1 shows the definition of the study samples. Statistical significance was set to a p-value < 0.05. All calculations were performed using IBM SPSS software version 26. Graphics were made with R software version 3.6.3 (R Foundation for Statistical Computing, Vienna, Austria; www.​R-project.​org).

The FI showed a right-skewed distribution (Additional Fig. 2). We observed a median FI of 0.27 [IQR 0.06, 0.65]. Notably 528 of 701 participants (75.3%) were identified as frail (FI ≥ 0.2). Orthopedic/other surgical patients yielded on average higher FI values (median 0.28; Q1 0.21|Q3 0.36) and were more often identified as frail (80.5%) than cardiac surgery patients (median 0.23; Q1 0.18|Q3 0.30; 68.2% frail). Median FI values were higher for women (median 0.28; Q1 0.21| Q3 0.37) than for men (median 0.25; Q1 0.18 | Q3 0.32) with 79.9% of women identified as frail versus only 71.1% of men (Table 3). The following nine deficits were identified among the top ten contributors to the FI for all patients, independently of the type of surgery: arterial hypertension, cardiovascular disease and osteoarthritis as comorbidities; taking ≥ 5 medications; reporting physical limitation by climbing several flight of stairs and/or performing moderate activities; having little levels of energy, subjective memory impairment or decreased hearing (Additional Table 3).

We observed 134 incident delirium events among frail subjects representing 25.4%, while only 31 deliriums could be detected among non-frail (17.9%). Compared to non-frail, frail participants were older, more often female, less educated, showed lower alcohol consumption and were more likely to live alone. Poorer physical functioning within the frail group was noticed with a higher rate of falls, lower levels of mobility, lower Barthel Index and worse results in TUG and handgrip strength (if available). Comparing comorbidities frail participants were diagnosed more often with arterial hypertension, cardiovascular disease, chronic lung disease, dementia, gastrointestinal ulcer, osteoarthritis, urinary incontinence, tumor, diabetes mellitus and liver disease than non-frail ones. Frail participants had lower visual and hearing abilities, lower GFR and higher rate of polypharmacy, anemia and higher BMI. In general, frail participants rated their health lower in the SF-12 than fit participants (Table 2).

Model 1 showed a statistically significant association between frailty and POD. Frail participants had higher odds for POD compared to non-frail (OR 1.66 [95% CI 1.06, 2.59]), where a 0.1 increment of the FI was associated with an OR of 1.32 [95% CI 1.12, 1.57] after adjustment for age and sex. After further adjustment in Model 2 including education, smoking, alcohol-score and type of surgery frailty showed a higher OR for POD (OR 2.14 [95% 1.33, 3.44], with an OR of 1.57 [95% CI 1.30, 1.90] for a 0.1 increase of the FI. The highest c-statistic (AUC) of 0.722 was seen in Model 2 with frailty as a continuous variable (Table 4).

The observed association between frailty and POD remained when performing the secondary analysis excluding those with a length of stay < 7 days and missing data on postoperative delirium at 2-months follow-up (n = 29) with a total of 672 participants without any sign for introduction of bias (Additional Table 4). We were able to build a 32-items FI including functional parameters such as TUG and handgrip strength for 517 of 701 participants. When compared to the study population this subpopulation was slightly younger with a higher ratio of men, cardiac surgery patients, lower rate of polypharmacy and less reported falls (Additional Table 5). No relevant differences in the odds ratios of Model 2 as well as in the distribution between the 30-items and the 32-items FI could be detected (Table 5, Additional Fig. 3). After adding parameters for muscle strength to the 30-items FI the OR for those frail (FI ≥ 0.2) went from 1.95 [95% CI 1.17, 3.25] to 1.80 [95% CI 1.11, 2.91] after adjustment for age, sex, education, smoking, alcohol consumption and type of surgery. Furthermore, no improvement was observed in model’s discrimination (both c-statistics 0.69). As a continuous variable, a 0.1 increment of the FI including parameters of muscle strength entailed an OR of 1.66 [95% CI 1.31, 2.09] compared to an OR of 1.61 [95% CI 1.28, 2.02] when using the 30-items FI, without any improvement in the discrimination (both c-statistic 0.72).

Our analysis included a large number of participants (n = 701), had a multicenter setting (5 medical centers) and evaluated different surgical types (cardiac and non-cardiac), allowing a more differentiated analysis. Moreover, two approaches helped to assess precisely postoperative delirium prevalence: I-CAM, a modified version of the commonly used CAM, and a Chart Review, which allowed to detect any delirium that might have been missed by I-CAM. Combining those two methods helped to keep the rate of missed delirium with its fluctuating course low. The proposed 30-items FI fulfills the requirement of multidimensionality as proposed by Searle et al. [16]. Because of missing baseline data we were able to build the FI only for 78.0% of the study population. Those excluded were noted to be more often living alone, to have a higher prevalence of falls in the last three months, with less prevalence for hypertension, visual or bilateral auditory impairments, and good functionality according to the Barthel Index (Additional table 6). Nevertheless, and although FI items are considered with the same weight in the model of accumulation of deficits, the observed distribution of the FI, its sex-specific variability along ages as well as the observed estimates for the association with delirium and the discriminatory power of the models, all of them consistent with the available literature, can be seen as indicators for a good construct of the index for this study population.

Unfortunately, only 57.5% of the population had data for functional physical parameters addressing muscle strength, known to be a good predictor for different outcomes and a surrogate for physical frailty in older adults [37]. Therefore we were able to evaluate the predictive value of a 32-item FI including TUG and handgrip strength only in a subsample of 517 patients. This secondary analysis within this subsample showed no further improvement in model’s discrimination after adding them, making the 30-item FI more feasible during preoperative evaluation since it contains mostly routinely gathered data not requiring the performance of extra time-consuming tests. Moreover, it can also be used for bed-bound patients (e.g. for hip surgery or fracture of femoral neck), which is a huge benefit compared to frailty assessments using mobility variables. In addition, and contrary to the clinical frailty scale, which captures a time frame [38], a prior clinical interaction with the patient is not needed in order to build a FI based on patient’s characteristics at the time of admission reinforcing its utility in surgical wards. To which extent our FI would be able to describe a relationship between frailty and delirium in geriatric or internal medicine settings could and should be analysed in further studies.