In this study reflecting real life practice in Germany, we examined rivaroxaban use in a cohort of more than 400 patients. We also explored an approach to investigate drug utilization for a drug which is started in hospital even though in-hospital prescribing information with a few exceptions is not available in the database providing data for this study.
This approach was based on several considerations: (i) during the study period, approval was limited to HR and KR thus a related in-hospital surgical procedure providing an exact date was likely; (ii) though thromboprophylaxis after KR or HR is recommended for only 14 days or 5 weeks, respectively, the tendency towards shorter stays in hospital after surgery  probably leads to a continuation of treatment in the outpatient setting where rivaroxaban use can be identified and its use extrapolated to the inpatient setting; (iii) as in Germany with hospital discharge only a very limited supply of medication e.g. to cover the weekend is given to the patient  an outpatient prescription within a few days after discharge could be expected; (iv) a change of treatment from subcutaneous injections of LMWH or fondaparinux in hospital to rivaroxaban after discharge was unlikely since during the time of the study rivaroxaban increased the costs for thromboprophylaxis compared to e.g. the LMWH enoxaparin or the indirect factor Xa inhibitor fondaparinux in the outpatient setting .
During the study period, for more than 91 % of the patients receiving an outpatient prescription of rivaroxaban, a preceding orthopaedic or surgical procedure could be identified qualifying for an in-hospital start of treatment. Overall, more than 82 % of the rivaroxaban treatment episodes were found to be on-label. Half of the remaining treatment periods were referred to other orthopaedic and surgical interventions including a small proportion of fracture surgery. Less than 3 % of the episodes were allocated to the cardiovascular indications approved in 2011 and 2012 and for 6 % no indication could be determined by our algorithm examining the 90 days preceding the first rivaroxaban prescription. However, extending the time interval to 180 days before and to 90 days following the first rivaroxaban prescription decreased the number of treatment episodes with no plausible indication to less than 4 %. Especially, including the 90 days following this first prescription yielded additional on-label indications. This might imply that rivaroxaban in these instances was prescribed prior to surgery and might partly be explained by office-based physicians with hospital affiliations conducting inpatient surgeries while prescriptions are organized by the medical practice and given to the patient before the intervention. The relatively high proportion of outpatient rivaroxaban treatment initiated by orthopaedic and other surgeons supports this assumption. Among those episodes where a potential indication for rivaroxaban use could be determined, 88 % were found to be on-label.
Overall, the recommendations on the duration of rivaroxaban use were not followed in most patients. In nearly three quarters of episodes, treatment exceeded the advised durations. The prolonged use which was mainly observed following KR might have been influenced by other guidelines such as the American College of Chest Physicians (ACCP) Evidence-Based Clinical Practice Guidelines published in 2008 (8th edition) suggesting that thromboprophylaxis for these patients be extended up to 35 days after surgery . Additionally, physicians might have chosen a prolonged treatment based on a patient’s clinical condition affecting post-surgery mobilization.
On the other hand, treatment was shorter than recommended in about one quarter of episodes putting patients at potential risk of thromboembolic events. These findings mainly applied to treatment following HR. Results of a similar magnitude were also reported from the multinational Global Orthopaedic Registry (GLORY) revealing that of those who received recommended forms of VTE prophylaxis after HR or KR the duration was shorter than recommended in approximately one quarter of the patients .
No patients were found to be younger than 18 years at cohort entry or to be pregnant during the study period. However, in several patients, diagnoses indicating liver or renal impairment were detected, which decreased when narrowing the time window to 180 days prior cohort entry and during rivaroxaban treatment. Given the high potential of changes in liver and renal function over time, which was reflected by the two time windows examined, fewer patients may have met the criteria of impairment of liver or renal function during the short treatment period. Additionally, though GePaRD does not provide laboratory data necessary for calculation of the impairment’s severity none of those patients for whom a stage of chronic renal failure could be deduced from the ICD codes fulfilled the severity of renal impairment described as contraindications and precautions.
Though examining potentially interacting drugs beyond those stated in the SPC, prescribing of medications with possible pharmacokinetic interactions during rivaroxaban treatment was rare and included none of the drugs not recommended for concomitant use. On the contrary, nearly half of the patients received NSAIDs which may interact via pharmacodynamic mechanisms. However, these drugs constitute plausible co-medications used to reduce post-operative pain and inflammation . NSAID prescriptions were slightly lower compared to a Dutch study which reported 52 % of patients receiving NSAIDs after HR or KR  and substantially lower than has been reported by an analysis of rivaroxaban clinical trials where over 70 % of patients with HR or KR concomitantly used NSAIDs . Since these studies supposedly included also NSAIDs administered in hospital [17, 18] these differences might be referred to a higher use in the first days after surgery. Prescribing of LMWH or fondaparinux during an estimated rivaroxaban treatment may indicate a change of therapy for example because of side effects. As VTE prophylaxis has been traditionally conducted with subcutaneous injections, insufficient knowledge about the new oral anticoagulants might have contributed to these findings as well. Since no prescribing of VKA was found in temporal relationship with rivaroxaban treatment there was no hint that rivaroxaban was used as bridging therapy in patients treated with VKA before or after the surgical intervention. As this study was not designed as a safety analysis, we did not examine whether these concomitantly prescribed drugs resulted in adverse effects.
Strengths of this study are the size of the underlying population and the lack of non-response due to the nature of administrative data . Determination of drug therapy based on pharmacy dispensing data is considered the gold standard as recall bias can be ruled out and information is precise in time and dose . By reconstructing and including inpatient treatment, we were able to depict the complete thromboprophylaxis following orthopaedic surgery for these patients while a restriction to outpatient prescriptions would have led to an underestimation of treatment time.
Limitations are mainly attributable to the nature of the administrative data. GePaRD does not include medication bought over the counter, thus an underestimation of e.g. NSAIDs is likely. Another shortcoming of our study was that it did not include a review of individual patient files which for data protection reasons is not feasible in Germany. So, although the respective inpatient procedures could be referred to an exact date and effective VTE prophylaxis is reported to be standard care in surgical wards in Germany , we had no possibility to verify that in patients receiving outpatient prescriptions of rivaroxaban thromboprophylaxis was actually started in hospital on the day of surgery.
A further limitation was that the duration of rivaroxaban treatment had to be estimated based on the prescribed package sizes since GePaRD does not provide the intended length of treatment. When rivaroxaban was introduced in Germany, packages of 10 or 30 tablets were available; an additional package size of 5 tablets was marketed later. As co-payment is required per package, patients might have been prescribed larger packages and told to stop earlier which in our study would have led to an overestimation of treatment time. Based on the recommended treatment durations and the package sizes available, it is likely that this scenario applied more often to those patients receiving rivaroxaban following knee replacement.
On the other hand, an underestimation of treatment duration could have resulted from observation periods being censored by either the end of the study or patients being hospitalized leading to inpatient rivaroxaban treatment. Overall, 15 % of episodes were censored; however, this applied to only 6 % of the episodes found to be shorter than recommended.
As no direct linkage is possible between prescriptions and corresponding indications, misclassification cannot be ruled out; however, the examined hospital procedures provided detailed information allowing for a distinction between rivaroxaban use following labelled and non-labelled surgeries. By adding information from diagnoses during the same hospital stays, we were able to differentiate between elective and fracture surgery as well. Given the high proportion of on-label indications examined it seems unlikely that misclassification is of great importance.
In conclusion, our study did not identify important off-label use of rivaroxaban, apart from that of extended treatment duration which might partly result from the estimation of treatment duration based on the package size of the prescription. Additionally, given the comparatively short recommended treatment durations in both indications, our study might have missed patients who were treated according to recommendations in hospital and during a possible stay in a rehabilitation clinic without receiving any outpatient rivaroxaban prescriptions.
Based on several assumptions, our study also provides an example of reconstructing inpatient drug use in a healthcare database which does not contain prescription information in hospital when treatment is continued in the outpatient setting. This approach requires that (i) the indication of the drug of interest and thus the start of in-hospital treatment can be specifically linked to an operation or procedure code with an exact date, (ii) a change of treatment between the in- and outpatient setting is unlikely and (iii) especially for short-time treatment only a small gap between hospital discharge and the first outpatient prescription can be expected. This approach might be useful for DUS dealing with the problem of drug use starting in hospital. However, possible limitations resulting from these assumptions should be considered carefully when interpreting the results especially when estimating treatment durations.