Few studies used subjective COC measures [ 53, 55, 68]. While these patient-reported measures are more susceptible to bias than objective COC measures, subjective measures are a valuable supplement to objective measures relying on claims data. Overall, our findings on utilized measures of COC are consistent with other studies showing that objective COC measures referring to relational continuity are most commonly used [ 24, 26, 33]. where each indicates the importance of the substructures with a radius of t. The final representation of a bond e i→ j, which captures the substructure information with different radii, is given by the weighted sum of bond-level hidden features across all steps according to the following: TWOSIDES is constructed by Zitnik et al. 39 after filtering and preprocessing the original TWOSIDES dataset. 40 It includes 645 drugs with 963 interaction types and 4 576 287 DDI tuples. As against the DrugBank dataset, these interactions are at the phenotypic level ( i.e., headache, pain in the throat, and others) rather than metabolic. The negative samples are generated by a procedure the same as the DrugBank dataset. 3.2 Experimental setup We compared the proposed SA-DDI with state-of-the-art methods, namely, DeepCCI, 26 MR-GNN, 29 SSI-DDI, 28 GAT-DDI, 30 and GMPNN-CS. 30 These baselines only consider chemical structure information as input and can work in both warm and cold start scenarios. The parameter settings for MR-GNN, SSI-DDI, and GMPNN-CS are consistent with their published source codes. As the source codes for DeepCCI and GAT-DDI are not provided, we implemented them with parameters recommended by the papers. 26,30 To investigate how the D-MPNN, substructure attention and substructure–substructure interaction module improve the model performance, we also consider the following variants of SA-DDI:

Rankin A, Cadogan CA, Patterson SM, Kerse N, Cardwell CR, Bradley MC, et al. Interventions to improve the appropriate use of polypharmacy for older people. Cochrane Database Syst Rev. 2018;9:CD008165. https://doi.org/10.1002/14651858.CD008165.pub4. cThe value is significantly different from the value for the corresponding control at a P of <0.05.Chen H-M, Tu Y-H, Chen C-M. Effect of continuity of care on quality of life in older adults with chronic diseases: a meta-analysis. Clin Nurs Res. 2017;26:266–84. https://doi.org/10.1177/1054773815625467. Fig. 7 Quantitative analysis of the substructure attention mechanism. (a) The relationship between accuracy and the number of iterations for the SA-DDI and SA-DDI_noSA. (b) The relationship between the F1-score and the number of iterations. (c) The distribution of substructure attention scores for the 10 iterations/steps SA-DDI in the DrugBank dataset. (d) The improvement of accuracy by increasing the number of iterations from 1 to 6 for SA-DDI_noSA. Non-linear mixed effect modelling quantitatively estimates the magnitude of unexplained variability in the population of interest. Identification of factors contributing to this variability and exploration of their relationship to exposure is an important component of the population pharmacokinetic approach and can be used to support dosing recommendations. For example, your primary phone number might be 01202 551000, and you ask for a range of 20 direct-dial-in numbers. Your provider would therefore issue a range such as this: Hajjar ER, Hanlon JT, Sloane RJ, Lindblad CI, Pieper CF, Ruby CM, et al. Unnecessary drug use in frail older people at hospital discharge. J Am Geriatr Soc. 2005;53:1518–23. https://doi.org/10.1111/j.1532-5415.2005.53523.x.

To the best of our knowledge, this study is the first to compare the mitochondrial and metabolic effects of five major NRTIs and three of their combinations in an in vivo murine model. NRTIs were added to the drinking water in daily doses corresponding to human therapeutic doses per body area. Age was found to impact the oral clearance to a statistically significant degree. CL/ F decreased by 0.28 L/hr per year of increasing age, corresponding up to a 40% higher exposure in a 75-year-old subject compared with the population mean of 44 years. The age effect on CL/ F cannot be explained by the fact that renal clearance decreases with age, since the renal clearance of vortioxetine is negligible. Vortioxetine is mainly metabolised through CYP2D6 and to a lesser degree through CYP2C19, but we are not aware of any age dependency (in the adult population) of CYP2D6 or CYP2C19 human activity in the literature. A relationship between creatinine clearance and clearance of many drugs eliminated via non-renal routes has previously been reported 13 , 14. A recent review of all new drug applications approved by the US Food and Drug Administration between 2003 and 2007 showed that of 23 non-renally excreted molecules whose pharmacokinetics had been studied in subjects with impaired kidney function, 13 displayed altered pharmacokinetics and six had alterations important enough to justify dose adjustments 15 , 16. In addition, a review by Touchette and Slaughter 13 reported that there exists some evidence, although it is inconclusive that the activity of the CYP2D6 isozyme is reduced with renal impairment. In the present population pharmacokinetic analysis, creatinine clearance was not a significant covariate in the final model, but could be because the inclusion criteria for the studies limit the range of creatinine clearance. The apparent age dependency of oral clearance may in fact be an effect of decreasing kidney function with age. However, the modest impact of age on the exposure of vortioxetine is not considered to be clinically relevant. Heider D, Matschinger H, Meid AD, Quinzler R, Adler J-B, Günster C, et al. The impact of potentially inappropriate medication on the development of health care costs and its moderation by the number of prescribed substances. Results of a retrospective matched cohort study. PLoS ONE. 2018;13:e0198004. https://doi.org/10.1371/journal.pone.0198004.Subjective measures of COC were used by three studies [ 65, 66, 69] (Table 2). In particular, patients were asked if they have a regular physician [ 65], whether they usually see the same physician [ 69], or whether they experienced a gap in care coordination [ 66]. These COC measures were treated as binary variables (yes vs no) (Tables S1 and S2, see ESM). Overall, a combination of the different types of COC measures was used by three studies [ 48, 57, 69]. 3.2.2 Operationalization of Polypharmacy Knowledge graph-based methods 16–23 represent biomedical data as graphs and use different graph-specific methods, such as label propagation, 20 matrix factorization, 21,23 and graph auto-encoders, 18 to analyze them. The advantage of knowledge graph-based methods is that the model performance can be boosted by external biomedical knowledge. However, these approaches cannot be generalized to drugs in the early development phase, because the only available information at that time is chemical structure. 18,20,24,25 Robles S, Anderson GF. Continuity of care and its effect on prescription drug use among Medicare beneficiaries with hypertension. Med Care. 2011;49:516–21. https://doi.org/10.1097/MLR.0b013e31820fb10c.

a) A directed message passing neural network (D-MPNN) 32 equipped with a novel substructure attention mechanism was presented to extract flexible-sized and irregular-shaped substructures. In SA-DDI, different scores determined by the substructure attention mechanism were assigned to substructures with different radii ( i.e., different receptive fields). The weighted sum of substructures centering at an atom with different radii results in a size-adaptive molecular substructure, as shown in Fig. 2. The substructure attention was also expected to assign a lower score to a substructure from a higher level to prevent over-smoothing. 33

inferred metabolic status was highly correlated with CL/ F in the first forward-inclusion step, with a drop of 72 points in OFV; but this relationship was not further evaluated in the next covariate model step due to over-parameterisation. Hanlon JT, Fillenbaum GG, Schmader KE, Kuchibhatla M, Horner RD. Inappropriate drug use among community-dwelling elderly. Pharmacotherapy. 2000;20:575–82. https://doi.org/10.1592/phco.20.6.575.35163.