Estimating the impact of physician risky-prescribing on the network structure underlying physician shared-patient relationships

Social network analysis and shared-patient physician networks have become effective ways of studying physician collaborations. Assortative mixing or "homophily" is the network phenomenon whereby the propensity for similar individuals to form ties is greater than for dissimilar individuals. Motivated by the public health concern of risky-prescribing among older patients in the United States, we develop network models and tests involving novel network measures to study whether there is evidence of geographic homophily in prescribing and deprescribing in the specific shared-patient network of physicians linked to the US state of Ohio in 2014. Evidence of homophily in risky-prescribing would imply that prescribing behaviors help shape physician networks and could inform interventions to reduce risky-prescribing (e.g., should interventions target groups of physicians or select physicians at random). Furthermore, if such effects varied depending on the structural features of a physician's position in the network (e.g., by whether or not they are involved in cliques - groups of actors that are fully connected to each other - such as closed triangles in the case of three actors), this would further strengthen the case for targeting of select physicians for interventions. Using accompanying Medicare Part D data, we converted patient longitudinal prescription receipts into novel measures of the intensity of each physician's risky-prescribing. Exponential random graph models were used to simultaneously estimate the importance of homophily in prescribing and deprescribing in the network beyond the characteristics of physician specialty (or other metadata) and network-derived features. In addition, novel network measures were introduced to allow homophily to be characterized in relation to specific triadic (three-actor) structural configurations in the network with associated non-parametric randomization tests to evaluate their statistical significance in the network against the null hypothesis of no such phenomena. We found physician homophily in prescribing and deprescribing in both the state-wide and multiple HRR sub-networks, and that the level of homophily varied across HRRs. We also found that physicians exhibited within-triad homophily in risky-prescribing, with the prevalence of homophilic triads significantly higher than expected by chance absent homophily. These results may explain why communities of prescribers emerge and evolve, helping to justify group-level prescriber interventions. The methodology could be applied to arbitrary shared-patient networks and even more generally to other kinds of network data that underlies other kinds of social phenomena.

Stability and spread: A novel method for quantifying transitions within multivariate binary time series data

We present a novel method for quantifying transitions within multivariate binary time series data, using a sliding series of transition matrices, to derive metrics of stability and spread. We define stability as the trace of a transition matrix divided by the sum of all observed elements within that matrix. We define spread as the number of all non-zero cells in a transition matrix divided by the number of all possible cells in that matrix. We developed this method to allow investigation into high-dimensional, sparse data matrices for which existing binary time series methods are not designed. Results from 1728 simulations varying six parameters suggest that unique information is captured by both metrics, and that stability and spread values have a moderate inverse association. Further, simulations suggest that this method can be reliably applied to time series with as few as nine observations per person, where at least five consecutive observations construct each overlapping transition matrix, and at least four time series variables compose each transition matrix. A pre-registered application of this method using 4 weeks of ecological momentary assessment data (N = 110) showed that stability and spread in the use of 20 emotion regulation strategies predict next timepoint affect after accounting for affect and anxiety's auto-regressive and cross-lagged effects. Stability, but not spread, also predicted next timepoint anxiety. This method shows promise for meaningfully quantifying two unique aspects of switching behavior in multivariate binary time series data.

Labour Market Dynamics and Worker Flows in India: Impact of Covid-19

Tracking and analyzing the labour market dynamics at regular, frequent intervals is critical. However, this was not possible for India, a large emerging economy with a significant population undergoing demographic transition, due to a paucity of data. We use the new dataset Centre for Monitoring Indian Economy (CMIE)-Consumer Pyramids Household Survey (CPHS) and use a panel to create Labour Flow Charts and Transition Matrices for India from January 2019 to December 2021. To the best of our knowledge, this is the first time these were created for India. We then use that to look at the impact of Covid-19 on the Indian labour market. We not only look at transitions between employment, unemployment and out of labour force, but also across types of employment-full-time and part-time. The rich data also allows us to consider heterogeneity in the labour market and look at the differential impact of the pandemic across different education groups and gender. From the labour flow charts and transition probabilities, we find that while all groups have been impacted, the magnitude of the impact is different across groups. The recovery is also uneven, and the extent depends on education levels. Further, we do an event study analysis to examine the likelihood of getting a full-time job across different educational and gender groups. Men, on average, enjoy a higher likelihood of getting a full-time job than women. The likelihood coefficients also go up with increasing educational qualifications. Looking at skill heterogeneity, while the likelihood of getting a full-time job either goes down for most groups during the pandemic or the change is minuscule, strikingly it goes up for those with no education, for both men and women. The likelihood coefficients remain elevated for men even after the restrictions are removed, and that for women reverts to the level seen before the pandemic. Finally, this paper provides a way to continuously monitor the dynamics of the labour market as data is released in the regular intervals in the future, which would be of great value for researchers and policymakers alike.

Resting state fMRI based multilayer network configuration in patients with schizophrenia

Novel methods for measuring large-scale dynamic brain organisation are needed to provide new biomarkers of schizophrenia. Using a method for modelling dynamic modular organisation (Mucha et al., 2010), evidence suggests higher 'flexibility' (switching between multilayer network communities) to be a feature of schizophrenia (Braun et al., 2016). The current study compared flexibility between 55 patients with schizophrenia and 72 controls (the COBRE Dataset). In addition, novel methods of 'between resting state network synchronisation' (BRSNS) and the probability of transition from one community to another were used to further describe group differences in dynamic community structure. There was significantly higher schizophrenia group flexibility scores in cerebellar (F (1124) = 9.33, p (FDR) = 0.017), subcortical (F (1124) = 13.14, p (FDR) = 0.005), and fronto-parietal task control (F (1124) = 7.19, p (FDR) = 0.033) resting state networks (RSNs), as well as in the left thalamus (MNI XYZ: -2, -13, 12; F(1, 124) = 17.1, p (FDR) < 0.001) and the right crus I (MNI XYZ: 35, -67, -34; F (1, 124) = 19.65, p (FDR) < 0.001). Flexibility in the left thalamus reflected transitions between communities covering default mode and sensory-somatomotor RSNs. BRSNS scores suggested altered dynamic inter-RSN modular configuration in schizophrenia. This study suggests less stable community structure in a schizophrenia group at an RSN and node level and provides novel methods of exploring dynamic community structure. Mediation of group differences by mean time window correlation did however suggest flexibility to be no better as a schizophrenia biomarker than simpler measures and a range of methodological choices affected results.

A continuous-time multistate Markov model to describe the occurrence and severity of diarrhea events in metastatic breast cancer patients treated with lumretuzumab in combination with pertuzumab and paclitaxel

PURPOSE

To inform lumretuzumab and pertuzumab dose modifications in order to decrease the incidence, severity, and duration of the diarrhea events in metastatic breast cancer patients treated with a combination therapy of lumretuzumab (anti-HER3) in combination with pertuzumab (anti-HER2) and paclitaxel using quantitative clinical pharmacology modeling approaches.

METHODS

The safety and pharmacokinetic (PK) data from three clinical trials (lumretuzumab monotherapy n = 47, pertuzumab monotherapy n = 78, and the combination therapy of lumretuzumab, pertuzumab and paclitaxel n = 35) were pooled together to develop a continuous-time discrete states Markov model describing the dynamics of the diarrhea events.

RESULTS

The model was able to capture the time course of different severities of diarrhea reasonably well. The effect of lumretuzumab and pertuzumab was well described by an E_max function indicating an increased rate of transition from moderate to mild or more severe diarrhea with higher doses. The concentration needed to trigger or worsen diarrhea episodes was estimated to be 120-fold lower in combination therapy compared to monotherapy, suggesting strong synergy between the two monoclonal antibodies. The prophylactic effect of loperamide in a subset of patients was also well captured by the model with a clear tendency to reduce the occurrence of diarrhea events.

CONCLUSIONS

This work shows that PK-toxicity modeling provides insight into how the severity of key adverse events evolves over time and highlights the potential use to support decision making in drug development.

Implications of iterative communication for biological system performance

The performance of integrated biological systems can often be described by the behavior of component subunits: the proportion of subunits performing an activity, and the rate of recruitment to the activity, can be relevant to system performance. We develop a model for activation of subunits (receivers) to a task when activation requires repeated signals (iterative communication). The model predicts how system performance will be affected by the parameters of iterative communication. Receiver activation is influenced by the frequency of stimulation, by forgetting about past interactions, and by the number of stimuli needed to activate the receivers. These parameters, along with the probability of activated receivers returning to a de-activated state, modulate the system-wide time course of activation and the steady-state proportion of activated receivers. Parameters can interact to affect system-wide activation, and multiple parameter combinations can yield similar patterns of activation. Group performance is less variable at higher stimulation frequencies and in systems with greater numbers of receivers. Biological constraints on iterative communication, such as time and energy costs, may limit the parameter values that are feasible for a given system. Iterative communication parameters may be subject to natural selection at the system (group) level because they affect system performance.

Comparing plant life histories using elasticity analysis: the importance of life span and the number of life-cycle stages

Recent studies have used transition matrix elasticity analysis to investigate the relative role of survival (L), growth (G) and fecundity (F) in determining the estimated rate of population increase for perennial plants. The relative importance of these three variables has then been used as a framework for comparing patterns of plant life history in a triangular parameter space. Here we analyse the ways in which the number of life-cycle stages chosen to describe a species (transition matrix dimensionality) might influence the interpretation of such comparisons. Because transition matrix elements describing survival ("stasis") and growth are not independent, the number of stages used to describe a species influences their relative contribution to the population growth rate. Reduction in the number of stages increases the apparent importance of stasis relative to growth, since each becomes broader and fewer individuals make the transition to the next stage per unit time period. Analysis of a test matrix for a hypothetical tree species divided into 4-32 life-cycle stages confirms this. If the number of stages were defined in relation to species longevity so that mean residence time in each stage were approximately constant, then the elasticity of G would reflect the importance of relative growth rate to λ. An alternative, and simpler, approach to ensure comparability of results between species may be to use the same number of stages regardless of species longevity. Published studies for both herbaceous and woody species have tended to use relatively few stages to describe life cycles (herbs: n=45, [Formula: see text]; woody plants: n=21, [Formula: see text]) and so approximate this approach. By using the same number of stages regardless of longevities, the position of species along the G-L side of the triangular parameter space largely reflects differences in longevity. The extent of variation in elasticity for L, G and F within and between species may also be related to factors such as successional status and habitat. For example, the shade-tolerant woody species, Araucaria cunninghamii, shows greater importance for stasis (L), while the gap-phase congener species, Araucaria hunsteinii, shows higher values for G (although values are likely to vary with the stage of stand development).