Objective: To describe the treatment patterns and time to next treatment (TTNT) in newly diagnosed multiple myeloma patients (MM) using a large-scale claims database in Japan.

Design: Cohort study

Methods: The patients with newly diagnosed MM from 2008 to 2015 were classified into two groups: age <65 years, and age ≥65 years. Specific regimens and general regimens were identified with a complex algorithm considering interval of no therapy, additional and discontinued agents. Correspondingly, TTNT between the first- and second-line were measured among non-transplant patients with Kaplan-Meier method.

Results: A total of 425 patients were eligible to participate in the analysis. The most common regimen for the treatment of MM was bortezomib-based regimens (52.9％ in the first-line, 28.2％ in later lines), followed by melphalan-prednisolone (27.1％ in the first-line, 12.9％ in later lines) and lenalidomide-based regimens (4.7％ in the first-line, 26.1％ in later lines). TTNT between the first- and second-line was 11.4 months and was seen to vary greatly with each regimen. A statistically longer TTNT was observed in subgroups of patients aged 65 years or over compared with patients aged younger than 65 years, but no statistical difference was found between conventional therapy and novel therapy.

Conclusion: Based on the data from the study, patients with MM were commonly treated with novel agent-based regimens, especially bortezomib-based regimens. Between the first- and second-line therapies a relatively short TTNT was observed, indicating that therapies in clinical practice poorly complied with treatment guidelines.

In the current review, the relationship between the reporting odds ratio (ROR) and the case-control study is addressed.

The proportional mortality ratio (PMR) obtained in the proportional mortality studies in the death registry cannot be regarded as the risk ratio (RR) in the cohort study, but, the mortality odds ratio (MOR) estimated by using deaths unrelated to the exposure as ‘controls’ can be regarded as the RR.

In 2004, Rothman et al proposed to estimate the ROR which can be regarded as the RR by using proper ‘control events’ in the spontaneous reports database. However, in the study conducted in Japan where the RORs estimated from 20 ‘control events’ were compared with the RRs obtained from many ‘drug use investigations’, the ROR vs RR plots were so diverse.

The author of the current review concludes that the study estimating the ROR in the spontaneous reports cannot be regarded as the case-control study as the case-control study should estimate the RR of the cohort study in the source population as the odds ratio (OR).

The ‘disproportionality measures’ like the ROR in the spontaneous reports database should be used primarily to detect the signals of the association between a drug and an adverse outcome. However, spontaneous reports can contribute to the characterization of the adverse drug reactions and to determining the causal relationship as well. The methods of signal detection are evolving and it is hoped that Japanese researchers contribute to their further developments.

Spontaneous reporting is an important source of information in pharmacovigilance. In Japan, the Japanese Adverse Drug Event Report database (JADER) was released in 2012, and this has led to numerous conference presentations and academic research papers that have reported the detection of signals based on the use of data mining methods. However, spontaneous reporting generally has certain limitations, including under-reporting, a lack of denominator information, and the effects of reporting bias, and these problems apply equally to JADER. The system of collecting spontaneous reports also influences the results obtained based on JADER analysis, as JADER in principle comprises serious adverse drug event reports and includes solicited reports. The detection of signals showing statistical significance does not necessarily imply a causal relationship between a particular drug and adverse events, and consequently, the cause of the signals detected requires careful interpretation. However, it has been pointed out that findings are sometimes accepted without considering the limitations.

For pharmaceutical companies, the Guidance on good pharmacovigilance practices Module Ⅸ and Guidance for Industry - Good Pharmacovigilance Practices and Pharmacoepidemiologic Assessment are available in the European Union and United States, respectively, for the purpose of signal management in pharmacovigilance activities. In contrast, there are limited resources to which researchers can refer when they publish scientific findings obtained using spontaneous reporting databases. To rectify this deficiency, we created a “checklist of important points to be noted during research that uses the data mining method in JADER (mainly signal detection by disproportionality analysis)” for the benefit of researchers using JADER. That study was supported by a Grant for Research Projects of the Japanese Society of Drug Informatics in 2017. In this article, we provide an overview of the checklist, with reference to the “Report of CIOMS Working Group Ⅷ： practical aspects of signal detection in pharmacovigilance,” which was used as a source when creating the checklist.