Discussion

In this longitudinal study of 12 111 cognitively normal participants, individuals with MBI had a 2.76-fold greater dementia incidence rate than those with no NPS, a difference not seen for those with NPS not meeting MBI criteria. Similar findings were observed in the subgroup of individuals with SCD. All MBI domains were associated with greater dementia incidence rate than no NPS, with MBI-psychosis having the highest hazard ratio and MBI-affective symptoms the lowest. The majority of dementia progressors with MBI progressed to Alzheimer's disease. These findings emphasise the importance of the core MBI criteria of symptom emergence and persistence in operationalising NPS-related dementia risk.

Several studies have reported on the association of global MBI with cognitive impairment and incident dementia. In a cross-sectional study of 499 participants with NPS assessed using the MBI-checklist (MBI-C),^{Reference Ismail, Aguera-Ortiz, Brodaty, Cieslak, Cummings and Fischer24} the MBI group demonstrated poorer performance in memory and executive function compared with no-MBI.^{Reference Kassam, Chen, Nosheny, McGirr, Williams and Ng12} A recent study that operationalised MBI using a single-timepoint NPS measure reported that MBI in cognitively normal participants was a significant predictor of progression to clinically diagnosed (hazard ratio 1.75) and neuropathologically confirmed Alzheimer's disease (hazard ratio 1.59).^{Reference Ruthirakuhan, Ismail, Herrmann, Gallagher and Lanctot25} In a longitudinal study of 9931 older adults without MCI or dementia, participants with MBI (captured with MBI-C) had a significantly worse baseline cognitive performance and greater decline in attention and working memory.^{Reference Creese, Brooker, Ismail, Wesnes, Hampshire and Khan13} An extension of this cohort with 8181 dementia-free older adults reported both cross-sectional and longitudinal associations between MBI and cognitive performance, with a moderating effect of gender. MBI and all its domains were significantly associated with cognitive decline, with a higher rate of decline in men than women.^{Reference Wolfova, Creese, Aarsland, Ismail, Corbett and Ballard26} In our sample of 12 111 cognitively normal older adults, participants with MBI had a 2.76-fold greater dementia incidence rate than those with no NPS prior to dementia onset. In contrast, no significant difference in dementia incidence rate was found for participants with NPS not meeting MBI criteria (NPS-not-MBI). While no moderating effect of gender was identified in our study, our findings are in line with the previous literature on the association of NPS with a greater rate of incident cognitive decline and dementia in cognitively unimpaired older adults.

Previous studies have also explored the association between NPS domains and incident dementia in cognitively normal older adults. A 14-year study of 4922 older men demonstrated that participants with depression (captured by the Geriatric Depression Scale), especially at the start of the follow-up period, were at greater dementia risk that those without.^{Reference Almeida, Hankey, Yeap, Golledge and Flicker6} A NACC study of 4517 participants explored the risk of progression to MCI or dementia across four classes of NPI-Q-assessed NPS. The greatest risk was reported for the multiple NPS class, followed by depressed and irritable classes. While each NPI-Q domain except elation was associated with a significantly greater risk of progression to MCI or dementia, delusions and hallucinations had the highest hazard ratio (4.5 and 3.4, respectively) despite relatively rare endorsement.^{Reference Leoutsakos, Forrester, Lyketsos and Smith27} Another study of 12 452 participants reported an association between baseline NPI-Q-measured psychosis (hazard ratio 3.6), agitation (hazard ratio 1.6) and affective symptoms (hazard ratio 1.5) and greater dementia risk, with psychosis having the greatest risk.^{Reference Liew8} Our MBI domain-level analyses revealed similar findings, with a 15.80-fold greater dementia incidence rate for MBI-psychosis, a 3.08-fold greater rate for MBI-impulse dyscontrol (encompassing agitation), and a 2.93-fold greater rate for MBI-affective dysregulation (encompassing depression).

With regard to MBI domain-level findings in cognitively normal participants, one NACC study using a single-timepoint NPS measure to operationalise MBI reported that all MBI domains were associated with Alzheimer's disease, with psychosis having the greatest effect (hazard ratio 6.49).^{Reference Ruthirakuhan, Ismail, Herrmann, Gallagher and Lanctot25} Other NACC studies also reported on specific MBI domains in mixed samples of cognitively normal individuals and those with MCI, with MBI-domain status determined based on symptom presence across two consecutive visits to meet the MBI symptom persistence criterion. In these studies, compared with No-NPS, the hazard ratio for dementia was 3.76 for MBI-psychosis,^{Reference Ismail, Ghahremani, Munir, Fischer, Smith and Creese28} 2.69 for MBI-apathy^{Reference Vellone, Ghahremani, Goodarzi, Forkert, Smith and Ismail29} and 1.76 for MBI-affective dysregulation.^{Reference Ebrahim, Ghahremani, Camicioli, Smith and Ismail30} In our MBI domain analyses, all domains were associated with greater dementia incidence; however, the hazard ratios were substantially higher than those of the previous literature, likely due to the methodological differences in capturing NPS. Our study operationalised the MBI symptom persistence criterion by capturing the presence of NPS across at least two-thirds of pre-dementia visits. In the three NACC studies on MBI psychosis, apathy, and affective dysregulation domains, the symptom persistence criterion was operationalised using two consecutive visits.^{Reference Ismail, Ghahremani, Munir, Fischer, Smith and Creese28–Reference Ebrahim, Ghahremani, Camicioli, Smith and Ismail30} Moreover, in these studies, all participants with a history of psychiatric disorders were excluded prior to assigning NPS group, thus primarily assessing the utility of the symptom persistence criterion. In our study, the exclusion criterion for history of psychiatric conditions was only applied to participants in the MBI group, better operationalising both core MBI criteria, identifying NPS that were more likely to represent sequelae of neurodegenerative disease, potentially at a preclinical or precognitive stage. Further, these domain-level studies evaluated mixed dementia-free samples of cognitively normal individuals and MCI, with interaction terms to determine cognitive status-specific hazard ratios. In the cognitively normal MBI-psychosis group, for example, the hazard ratio was 9.96 versus no-NPS, compared with 3.38 in MCI MBI-psychosis.^{Reference Ismail, Ghahremani, Munir, Fischer, Smith and Creese28} Similarly, for MBI-apathy, the hazard ratio in cognitively normal individuals was 5.91 versus 2.16 in MCI^{Reference Vellone, Ghahremani, Goodarzi, Forkert, Smith and Ismail29} and for MBI-affective dysregulation, the hazard ratio in cognitively normal individuals was 2.59 versus 1.50 in MCI.^{Reference Ebrahim, Ghahremani, Camicioli, Smith and Ismail30}

To our knowledge, the only study on the association of MBI-impulse dyscontrol and MBI-social inappropriateness with incident dementia reported adjusted hazard ratios of 2.02 and 2.25 for each domain, respectively, using a single-timepoint NPS assessment.^{Reference Ruthirakuhan, Ismail, Herrmann, Gallagher and Lanctot25} The present study used a more precise approach to operationalise both MBI core criteria of symptom emergence and persistence and demonstrated significant associations of these domains with incident dementia in cognitively normal older adults, but with a greater magnitude of effect (hazard ratio 7.86 for MBI-social inappropriateness, hazard ratio 3.08 for MBI-impulse dyscontrol). While social inappropriateness and its association with incident dementia is understudied, previous literature indicate deficits in social cognition in MCI^{Reference Henry, von Hippel, Thompson, Pulford, Sachdev and Brodaty31–Reference Teng, Lu and Cummings34} and dementia.^{Reference Hargrave, Maddock and Stone35,Reference Dermody, Wong, Ahmed, Piguet, Hodges and Irish36} Our study demonstrates that even in a state of objectively normal cognition, older adults can present with socially inappropriate behaviours that may be the first sign of an unrecognised neurodegenerative disease in advance of noticeable cognitive impairments. Similarly, symptoms of impulsivity and agitation are common in MCI.^{Reference Apostolova and Cummings37–Reference Copeland, Daly, Hines, Mastromauro, Zaitchik and Gunther39} In a cross-sectional study of 1377 dementia-free older adults, MBI-impulse dyscontrol was among the most prevalent domains, with frequencies of 17.2% in normal cognition, 28.7% in SCD and 33.8% in MCI, although assessed using a single-timepoint NPS measure.^{Reference Mortby, Ismail and Anstey40} Unfortunately, patients expressing socially inappropriate or impulsive behaviours are often misdiagnosed with a psychiatric disorder, without recognising that symptoms may represent early manifestations of an underlying neurodegenerative disease.^{Reference Desmarais, Lanctot, Masellis, Black and Herrmann41,Reference Cieslak, Smith, Lysack and Ismail42} Our findings demonstrate that in cognitively normal older adults, later-life emergent and persistent socially inappropriate and impulsive behaviours are significantly associated with greater dementia incidence rates. Earlier identification of these symptoms could identify treatment targets, potentially improving quality of life for patients and their caregivers.

In our analyses, for both the cognitively normal sample and the SCD subsample, across NPS groups, MBI had the highest percentage of participants progressing to dementia (Fig. 2). Among progressors across all NPS groups, the majority progressed to Alzheimer's disease dementia and only few had bvFTD, LBD or VaD, a result which may be surprising to some. However, given the high population prevalence of Alzheimer's disease relative to other dementias, the fact that 30% of Alzheimer's disease cases have a behavioural prodrome, and that defining the risk group in accordance with MBI criteria provides more specificity, this result is rather expected.

Despite fewer studies on NPS and SCD compared with MCI, evidence suggests an association between NPS and incident dementia in older adults with SCD. In a longitudinal study of 579 710 older adults, while SCD was significantly associated with greater dementia risk, the risk was greater still in the presence of depressive symptoms.^{Reference Lee, Kang, Lee, Kim, Kim and Yu43} Similar findings were reported in a longitudinal study of 2415 cognitively unimpaired older adults, in which the risk of dementia roughly doubled in participants with SCD and worry symptoms compared with SCD alone.^{Reference Jessen, Wiese, Bachmann, Eifflaender-Gorfer, Haller and Kolsch44} Furthermore, in a NACC study of 2769 cognitively unimpaired (CDR = 0) older adults, the combination of MBI and SCD was associated with a greater likelihood of CDR progression to ≥0.5 at 3 years compared with either MBI or SCD alone.^{Reference Ismail, McGirr, Gill, Hu, Forkert and Smith5}

In our primary analysis of 12 111 cognitively normal participants, compared with the reference group with no SCD and no NPS, the co-occurrence of MBI and SCD had the highest hazard ratio for incident dementia (hazard ratio 5.10, CI: 3.90–6.66) compared with when SCD was present with no NPS (hazard ratio 2.60, CI: 1.97–3.44) or with NPS-not-MBI (hazard ratio 2.30, CI: 1.85–2.86) (Supplementary Table 3). These findings are consistent with the previous 3-year study of CDR progression.^{Reference Ismail, McGirr, Gill, Hu, Forkert and Smith5} In our interaction analysis, the relative impact of MBI on dementia risk within the SCD stratum was lower (hazard ratio 1.96) than the impact of MBI in the cognitively normal stratum (hazard ratio 3.53). This is while the relative impact of NPS-not-MBI was 0.88 in SCD and 0.99 in cognitively normal individuals, indicating that NPS-not-MBI are not significant contributors to dementia risk over no NPS. These findings are consistent with several previous analyses in mixed dementia-free samples, with the same intuitive explanation for the interaction across cognitive strata.^{Reference Ismail, Ghahremani, Munir, Fischer, Smith and Creese28–Reference Ebrahim, Ghahremani, Camicioli, Smith and Ismail30,Reference Ghahremani, Wang, Chen, Zetterberg, Smith and Ismail45} In cognitively unimpaired participants without subjective memory complaints, cognitive status contributes little to risk, which is driven by behavioural status (namely MBI, as NPS-not-MBI did not differ from No-NPS). However, subjective memory complaints may indicate a more advanced disease stage than normal cognition. Correspondingly, the relative impact of behaviour on risk is less in SCD versus normal cognition, as cognitive symptoms also contribute to risk. Importantly, the absolute risk is greater in this group (hazard ratio 5.10 in SCD versus hazard ratio 3.53 in cognitively normal individuals), which represents an interaction of cognitive risk and behavioural risk.

The sensitivity analysis was also informative. In the SCD sample of 3485 participants, MBI had a greater hazard ratio for incident dementia (hazard ratio 1.99) than NPS-not-MBI (hazard ratio 0.92) and No-NPS (reference level; hazard ratio 1). These findings provide additional evidence that different approaches to incorporating NPS into dementia prognostication can result in significantly different estimates. The incorporation of natural history (symptom emergence in later-life and symptom persistence) into the predictor representing behavioural symptoms improves the prognostic power of these behavioural symptoms over what has traditionally been a cross sectional measure. Thus, adding MBI status to the traditional assessment of risk represented by cognitive status could improve the specificity of the risk estimate.