Ires Ghielen

71 Replication factor analysis in neuropsychiatric PD patients Methodological considerations Our initial plan to match the two patient groups on age and sex resulted in small sample sizes (n=139) without solving the differences in demographic and clinical variables. We therefore decided to perform our analyses with the largest sample sizes to retain the highest possible power. In network analysis, it is yet to be determined what is considered a satisfactory sample size [17]. It is suggested to perform network analysis on the largest samples since the networks are then estimated more accurately [15]. Relatively small sample sizes are a well-known issue in clinical research and with this it is considered challenging to gain an accurate estimation of a network [15], which was also the case in the present study. Both the low-anxious and high-anxious patient group showed an instable network, represented by the wide bootstrapped confidence intervals (figure 2), which limited us to draw reliable conclusions. In addition, we selected the patient groups on a total score of a questionnaire that was also directly represented as symptoms in the networks, which might have increased false positive associations known as Berkson’s bias [22]. To avoid false positive results as much as possible, we conducted tuned analyses to gain parsimonious networks, and we corrected for multiple comparisons. Obviously, the symptoms in our network are dependent on the items from the data sources (i.e. clinical assessments or questionnaires) on which they are based. Our network might therefore never be ‘complete’ (i.e. include all symptoms of the participating patients) and can involve different items that might measure the same symptoms, as is also discussed more in depth in the previous section. In addition, there might be other variables that influence the associations between items that were not accounted for, such as within-person variability and use of medication. We studied motor an anxiety symptoms in PD as static measures using a cross- sectional approach. Therefore, we were not able to determine how these symptoms interact over time, as is especially relevant in the context of response fluctuations during the day [6, 11]. Network analysis may be very useful to visualize fluctuations of motor, autonomic and neuropsychiatric symptoms over time, using high-frequent individual time-series data from ecological monitoring assessment tools. It would be interesting to apply this statistical method to these response fluctuations and investigate the interactions between motor and anxiety symptoms more in depth. This also enables the study of direct effects of medication and environmental factors on the network architecture over time. 4