The Drug Prescription Network: A System-Level View of Drug Co-Prescription in Community-Dwelling Elderly People

An introduction to network analysis for studies of medication use

BACKGROUND

Network Analysis (NA) is a method that has been used in various disciplines such as Social sciences and Ecology for decades. So far, NA has not been used extensively in studies of medication use. Only a handful of papers have used NA in Drug Prescription Networks (DPN). We provide an introduction to NA terminology alongside a guide to creating and extracting results from the medication networks.

OBJECTIVE

To introduce the readers to NA as a tool to study medication use by demonstrating how to apply different NA measures on 3 generated medication networks.

METHODS

We used the Norwegian Prescription Database (NorPD) to create a network that describes the co-medication in elderly persons in Norway on January 1, 2013. We used the Norwegian Electronic Prescription Support System (FEST) to create another network of severe drug-drug interactions (DDIs). Lastly, we created a network combining the two networks to show the actual use of drugs with severe DDIs. We used these networks to elucidate how to apply and interpret different network measures in medication networks.

RESULTS

Interactive network graphs are made available online, Stata and R syntaxes are provided. Various useful network measures for medication networks were applied such as network topological features, modularity analysis and centrality measures. Edge lists data used to generate the networks are openly available for readers in an open data repository to explore and use.

CONCLUSION

We believe that NA can be a useful tool in medication use studies. We have provided information and hopefully inspiration for other researchers to use NA in their own projects. While network analyses are useful for exploring and discovering structures in medication use studies, it also has limitations. It can be challenging to interpret and it is not suitable for hypothesis testing.

A Network Approach for the Study of Drug Prescriptions: Analysis of Administrative Records from a Local Health Unit (ASL TO4, Regione Piemonte, Italy)

In a Drug Prescription Network (DPN), each drug is represented as a node and two drugs co-prescribed to the same patient are represented as an edge linking the nodes. The use of DPNs is a novel approach that has been proposed as a means to study the complexity of drug prescription. The aim of this study is to demonstrate the analytical power of the DPN-based approach when it is applied to the analysis of administrative data. Drug prescription data that were collected at a local health unit (ASL TO4, Regione Piemonte, Italy), over a 12-month period (July 2018–June 2019), were used to create several DPNs that correspond to the five levels of the Anatomical Therapeutic Chemical classification system. A total of 5,431,335 drugs prescribed to 361,574 patients (age 0–100 years; 54.7% females) were analysed. As indicated by our results, the DPNs were dense networks, with giant components that contain all nodes. The disassortative mixing of node degrees was observed, which implies that non-random connectivity exists in the networks. Network-based methods have proven to be a flexible and efficient approach to the analysis of administrative data on drug prescription.

The similarity of inherited diseases (II): clinical and biological similarity between the phenotypic series

Background

Despite being caused by mutations in different genes, diseases in the same phenotypic series are clinically similar, as reported in Part I of this study. Here, in Part II, we hypothesized that the phenotypic series too might be clinically similar. Furthermore, on the assumption that gene mutations indirectly cause clinical phenotypes by directly affecting biological functions, we hypothesized that clinically similar phenotypic series might be biologically similar as well.

Methods

To test these hypotheses, we generated a clinical similarity network and a set of biological similarity networks. In both types of network, the nodes represent the phenotypic series, and the edges linking the nodes indicate the similarity of the linked phenotypic series. The weight of each edge is proportional to a similarity coefficient, which depends on the clinical phenotypes and the biological features that are shared by the linked phenotypic series, in the clinical and biological similarity networks, respectively.

Results

After assembling and analyzing the networks, we raised the threshold for the similarity coefficient, to retain edges of progressively greater weight. This way all the networks were gradually split into fragments, composed of phenotypic series with increasingly greater degrees of similarity. Finally, by comparing the fragments from the two types of network, we defined subsets of phenotypic series with varying types and degrees of clinical and biological correlation.

Conclusions

Like the individual diseases, the phenotypic series too are clinically and biologically similar to each other. Furthermore, our findings unveil different modalities of correlation between the clinical manifestations and the biological features of the inherited diseases.

General Practitioners Records Are Epidemiological Predictors of Comorbidities: An Analytical Cross-Sectional 10-Year Retrospective Study

Background. Comorbidity represents the co-occurrence of pathological conditions in the same individual, and presents with very complex patterns. In most cases, reference data for the study of various types of comorbidities linked to complex diseases are those of hospitalized patients. Such patients may likely require cure due to acute conditions. We consider the emerging role of EHR (Electronic Healthcare Records), and study comorbidity patterns in a general population, focusing on diabetic and non-diabetic patients. Methods. We propose a cross-sectional 10-year retrospective study of 14,958 patients and 1,728,736 prescriptions obtained from family doctors, and thus refer to these data as General Practitioner Records (GPR). We then choose networks as the tools to analyze the diabetes comorbidity patterns, distinguished by both prescription type and main patient characteristics (age, gender). Results. As expected, comorbidity increases with patients’ age, and the network representations allow the assessment of associations between morbidity groups. The specific morbidities present in the diabetic population justify the higher comorbidity patterns observed in the target group compared to the non-diabetic population. Conclusions. GPR are usually combined with other data types in EHR studies, but we have shown that prescription data have value as standalone predictive tools, useful to anticipate trends observed at epidemiological level on large populations. This study is thus relevant to policy makers seeking inference tools for an efficient use of massive administrative database resources, and suggests a strategy for detecting comorbidities and investigating their evolution.