Network structures and temporal stability of self- and informant-rated affective symptoms in Alzheimer's disease

Abstract

Background: Affective symptoms in Alzheimer's disease (AD) can be rated with both informant- and self-ratings. Information from these two modalities may not converge. We estimated network structures of affective symptoms in AD with both rating modalities and assessed the longitudinal stability of the networks.

Methods: Network analyses combining self-rated and informant-rated affective symptoms were conducted in 3198 individuals with AD at two time points (mean follow-up 387 days), drawn from the NACC database. Self-rated symptoms were assessed by Geriatric Depression Scale, and informant-rated symptoms included depression, apathy and anxiety questions from Neuropsychiatric Inventory Questionnaire.

Results: Informant-rated symptoms were mainly connected to symptoms expressing lack of positive affect, but not to the more central symptoms of self-rated worthlessness and helplessness. Networks did not differ in structure (p = .71), or connectivity (p = .92) between visits. Symptoms formed four clinically meaningful clusters of depressive symptoms and decline, lack of positive affect, informant-rated apathy and anxiety and informant-rated depression.

Limitations: The symptom dynamics in our study could have been present before AD diagnosis. The lack of positive affect cluster may represent a methodological artefact rather than a theoretically meaningful subgroup. Requiring follow-up lead to a selection of patients with less cognitive decline.

Conclusions: Informant rating may only capture the more visible affective symptoms, such as not being in good spirits, instead of more central and severe symptoms, such as hopelessness and worthlessness. Future research should continue to be mindful of differences between self- and informant-rated symptoms even in earlier stages of AD.

Keywords: Alzheimer's disease; Anxiety; Apathy; Depression; Network analysis; Neuropsychiatric symptoms.

Figure 1.

Network structures at baseline (A) and at follow-up (B). Orange nodes represent NPI-Q symptoms, as reported by an informant. Green nodes represent the twelve GDS items thought to assess depressive symptoms, and blue nodes correspond to the three GDS items related to apathy. Edges, or blue lines between the nodes, denote unique connections when conditioning for all other nodes in the network (van Borkulo et al., 2015), where thicker edges denote stronger connections. Blue circles around the nodes depict the degree of normalized correct classification, which is an index of predictability for binary data above what is trivially predicted by the relative probability of given condition (symptom present or not) irrespective of other nodes (Haslbeck & Waldorp, 2018). Layout of the network is averaged over the two visits, and reverse-scored items are indicated by >< in the legend.

Figure 1.

Network structures at baseline (A) and at follow-up (B). Orange nodes represent NPI-Q symptoms, as reported by an informant. Green nodes represent the twelve GDS items thought to assess depressive symptoms, and blue nodes correspond to the three GDS items related to apathy. Edges, or blue lines between the nodes, denote unique connections when conditioning for all other nodes in the network (van Borkulo et al., 2015), where thicker edges denote stronger connections. Blue circles around the nodes depict the degree of normalized correct classification, which is an index of predictability for binary data above what is trivially predicted by the relative probability of given condition (symptom present or not) irrespective of other nodes (Haslbeck & Waldorp, 2018). Layout of the network is averaged over the two visits, and reverse-scored items are indicated by >< in the legend.

Figure 2.

Standardized strength estimates of the nodes at baseline and at follow-up. Strength is a centrality measure for networks, which indicates the direct connectedness of a node (Epskamp et al., 2018). Strength values are the summary of edge weights connecting to a node, and these values were standardized for comparison in the figure.

Figure 3.

Community detection labelled networks using walktrap algorithm at baseline (A) and follow-up (B), where colors represent membership of communities. Note the discrepancies between a priori divisions in Figure 1 and this figure. Orange nodes represent the five reverse-scored items, interpreted as lack of positive affect. Blue nodes include depressive symptoms and symptoms that relate to decreasing capabilities, green nodes depict informant-rated apathy and anxiety, and the yellow node represents the one-node community of informant-rated depression.

Figure 3.

Community detection labelled networks using walktrap algorithm at baseline (A) and follow-up (B), where colors represent membership of communities. Note the discrepancies between a priori divisions in Figure 1 and this figure. Orange nodes represent the five reverse-scored items, interpreted as lack of positive affect. Blue nodes include depressive symptoms and symptoms that relate to decreasing capabilities, green nodes depict informant-rated apathy and anxiety, and the yellow node represents the one-node community of informant-rated depression.