The de-identified data were extracted from routinely collected data from a consecutive sample of community-dwelling older patients assessed by the Memory Service at Te Whatu Ora Counties Manukau, spanning 2013 to 2021. Approximately half of assessed patients were diagnosed with dementia, and half did not have dementia. The memory service primarily accepts referrals from primary care providers and some secondary care services, excluding individuals already residing in residential care. To meet the criteria for referral to the memory service, patients must have a primary concern related to subjective and/or objective cognitive decline.

Dementia diagnoses, including subtypes and severity, were established through clinical consensus during weekly multidisciplinary team meetings at the memory service. These diagnoses were based on a combination of clinical information and neuroimaging findings. The diagnostic criteria used for dementia followed DSM-5 guidelines,¹ while the severity of dementia was assessed using the Clinical Dementia Rating (CDR) criteria.¹⁹ The Confusion Assessment Method or CAM²⁰ is a questionnaire that screens for delirium. It can be completed within five minutes and is based on four features: acute onset and fluctuation course (i), inattention (ii), disorganised thinking (iii) and altered level of consciousness (iv). The delirium diagnosis requires the presence of features (i) and (ii) as well as either (iii) or (iv).

Routinely collected data extraction included age, gender, ethnicity, CAM scores and dates, hospitalisation data, care home placements and mortality data. We defined delirium as scoring positive on all four CAM questions at least once during any hospital admission after they were assessed for dementia. CAM was scored on routine clinical observation. Systematic delirium screening was introduced in the hospital for patients aged 65+ quite recently, so full coverage has not yet been achieved. Initial and updated diagnoses of dementia were routinely collected. Subsequent delirium episodes were considered after the updated dementia diagnosis for those who had it. For patients without dementia, given the characteristics of the data, we assumed they did not change in the diagnosis.

All data were de-identified before analyses. Patient ethnicities were categorised as NZ European, Māori, Pacific Islander, and other (although people of Chinese and Indian ethnicity are major ethnic groups in NZ, the numbers in our data did not warrant a separate category). We created four diagnostic categories regarding dementia and delirium: delirium only, dementia only, dementia and (at least one episode of) delirium, and no delirium/no dementia. Baseline characteristics between groups were compared using the chi-squared test, Fisher’s exact test, student t-test, Mann-Whitney U test, analysis of variance (ANOVA), and Kruskal-Wallis test. Logistic and multivariate Cox regression were carried out to determine the relationship between diagnostic category, mortality, care home placement and sociodemographic variables. A generalised additive model for location, scale and shape with generalised gamma response was used to analyse the association between diagnostic category and length of stay because it tolerates skewed data. We used a level of significance of 5%. All statistical analyses were made using statistical software R 4.3.1 version.²¹

Of the 798 patients who were hospitalised and screened at least once for delirium, 538 had dementia, 257 did not have dementia, and three did not receive a diagnosis due to incomplete assessment. Of the 795 who had a full dementia assessment and a delirium screen, 213 (27%) screened positive for delirium, and 178 (84%) of these also had dementia. A logistic regression was fitted to assess which sociodemographic and clinical characteristics were associated with delirium, see Supplementary Table 2. Due to their collinearity with number of delirium screens per person-year, total days in hospital per person-year and number of admissions per person-year were not included in the model. The number of delirium tests was included in the model to ensure that the occurrence of delirium did not have a confusing effect with the number of delirium screens since it was expected that patients with a greater number of screens would have a high probability of delirium.

We found that age and gender were not associated with delirium. On unadjusted analysis, we found that Māori were less likely to screen positive for delirium. In contrast, Pacific peoples were more likely to screen positive. Still, these findings were no longer statistically significant when they were adjusted for dementia status, gender, age, and number of delirium screens per person-year.

A diagnosis of dementia increased the likelihood of delirium three-fold. Compared to individuals without delirium, we found that each extra delirium screen increased the chance of being screened positive by 30%, and we adjusted for number of delirium screens per person-year, but this made no difference to the associations between dementia and delirium, adjusted OR=3.0 (95% CI:2.0-4.6).

We divided the screened sample into four diagnostic categories based on their initial dementia assessment and whether they had screened positive for delirium at least once during the period of follow-up: delirium only (n=35), dementia only (n=360), delirium and dementia (n=178), and no delirium/no dementia (n=222). Table 1 shows the association between diagnostic category, sociodemographic characteristics, and clinical outcomes. As might be expected, older patients were more likely to have a diagnosis of dementia and/or delirium.

There was an association between diagnostic category and ethnicity: Pacific patients had a higher frequency of delirium and/or dementia than other ethnic groups, whereas Māori had a lower frequency. The adjusted relative risk ratio of having dementia only or delirium superimposed on dementia, compared to those with neither, was 2.3 (p<0.001) and 2.9 (p<0.001) in Pacific people compared to NZ European. However, the adjusted risk ratio for delirium superimposed on dementia remained lower for Māori (RR=0.46, p=0.021) compared with NZ Europeans.

Table 2 presents the association between diagnostic category and length of stay, care home placement, and mortality over the nine-year follow-up period.

Length of stay:Supplementary Figure 1 shows the total length of stay per person-year by diagnostic category. When patients were discharged to a care home, this was associated with longer length of stay, so we adjusted the final model for care home status. After adjustment for age, sex, ethnicity and hospital discharge to care home, Table 2 shows that the mean total length of stay per person-year for patients with delirium was 2.6 times higher compared to patients without delirium or dementia (p<0.001) and 1.3 times higher in patients with both delirium and dementia (p=0.044) whereas having dementia only did not make a difference to length of stay.

Care home placement:Figure 1 shows the Kaplan–Meier survival curve for care home placement by diagnostic category. After adjustment for age, sex, and ethnicity, Table 2 shows that patients with dementia and/or delirium had two to four times the risk of care home admission compared to patients with neither diagnosis (p<0.001).

Mortality:Figure 1 shows the Kaplan–Meier survival curve for mortality by diagnostic category. Patients with delirium only had the lowest probability of survival, followed by those with delirium and dementia (p<0.001). After adjustment for age, sex, and ethnicity, patients with delirium (with or without dementia) had a three-fold higher risk of death, whereas patients with dementia only had a 70% higher risk of death compared to patients with neither diagnosis.

Ethnicity: We were unable to test for interaction between diagnostic category, ethnicity and outcomes due to small sample size.

The main strength of our study is that every patient had a thorough diagnostic assessment through the memory service, ensuring good case ascertainment of dementia. Due to the use of routinely collected data, we were also able to include everyone who had a delirium screen, whereas participants who are too unwell to consent to participate may be excluded in prospective studies. Along with other previous studies,^10,14 have shown that we can use routinely collected data to replicate the findings from similar prospective cohort designs,^{5,9,11–13,16} which, although not as methodologically rigorous, demonstrates the utility and cost-effectiveness of this study method.

There are three major limitations of the study. First is the incomplete ascertainment of delirium status as only half of the hospitalised patients were screened, despite hospital policy being that delirium screening should be conducted on all inpatients aged 65+. However, patients with dementia diagnoses in our sample were no more likely to be screened for delirium than those without dementia (p=0.71), which we believe demonstrates there was no selection bias that might have explained the association of delirium with dementia. Secondly, we defined delirium as a CAM score of 4, whereas, depending on which items are endorsed, some people with delirium will have a CAM score of 3; therefore, we may have underestimated the number of people with delirium in our sample. Moreover, CAM was scored on routine clinical observation; we did not score it based on answers to a structured cognitive assessment, as suggested by Inouye et al. (2001).²² The CAM has also yet to be validated in New Zealand, specifically in the culturally diverse populations in the study. Given cultural differences and especially linguistic difficulties, delirium screening tests may misdiagnose people of non-English speaking background. The third limitation is that we were unable to measure whether delirium was an independent predictor of adverse outcomes or whether its impact was mediated by physical comorbidity, functional status, and/or frailty, all of which are associated with length of stay, mortality, length of stay and care home placement. However, those previous studies that adjusted for physical comorbidity, functional status, and/or frailty could still show an independent effect of delirium.^9,12,13,15

In addition to measuring potential mediators/confounders in future studies, we recommend using the 4AT as the delirium screening tool, as this has been shown to have better diagnostic accuracy for delirium than the CAM.^23,24 Future research might also focus on interventions to prevent or reduce the impact of delirium, focusing on a culturally diverse population like New Zealand. Routine screening in hospitals has been shown to improve identification rates of delirium and dementia, and interventions in hospitals and care homes have been shown to be effective in preventing up to 40% of delirium (e.g., the Hospital Elder Life Program,²⁵).

CGP and SC designed the study. CGP, SY and BO extracted and de-identified the data. CGP carried out the statistical analysis. CRR supervised the statistical analysis. CGP, SC, and EG wrote the first draft of the manuscript. GD, DW, SY and CRR reviewed subsequent versions of the manuscript. All authors participated in the review of the final version of the manuscript and approved the publication.

This research was approved by the Health and Disability Ethics Committee (HDEC) reference: 17/NTB/191.

Authors declared no conflicts of interest.