Inaccuracy in meta‐analysis on safety of eleven drugs for the treatment of Parkinson's disease

We read with interest the article “Adverse effects produced by different drugs used in the treatment of Parkinson's disease: A mixed treatment comparison” by Li et al1 It is very meaningful that the authors performed a mixed treatment comparison on the topic of adverse effects of eleven drugs used to treat Parkinson's disease to inform clinical decision‐making. However, we have some concerns on the accuracy in data collection, extraction, and synthesis, which is a crucial issue for the determination of reliability of results. So I feel obliged to point out these mistakes.

First, we noticed that Li et al came a conclusion in the direct comparison results that “compared with levodopa, ropinirole presented with higher incidence of dyskinesia on PD (OR = 3.55, 95% CI = 1.76‐7.14)” (Section 3.2, Page 831). On the contrary, one of included study of this mixed treatment comparison by Watts et al2 showed that compared with increasing doses of levodopa, ropinirole prolonged release delayed the onset of dyskinesia in early PD patients. To search for the explanation of this contradiction, we checked the details in data collection, extraction, and synthesis of the present mixed treatment comparison. Finally, an error in the data extraction from the study by Rascol et al3 2000 was found for the direct comparison of B vs L in the incidence of dyskinesia (Line 31, Page 832, Table 2), with the data wrongly presented as L vs B. This error also occurs in other direct comparison results of B vs L, such as nausea, dizziness, constipation, and somnolence.

Second, data synthesis in the direct comparison results is incomplete. For example, Watts et al 2010 and Rascol et al 2000 both reported the incidence of dyskinesia, while the author only extracted the data of the Rascol et al 2000 (in Table 2). Similarly, Im et al4 and Brunt et al5 both reported the incidence of dizziness, but the author only extracted the data of 1 study(in Table 3) that is inconsistent with that of Im et al 2003 or Brunt et al 2002.

Third, the included literatures in this mixed treatment comparison are not fully in accordance with the inclusion criteria. We noticed that “patients diagnosed with PD that were over 50 years old” are one of the declared inclusion criteria, while in most of the included literatures, the recruited patients are aged over 30 or 40 years. Supposing that the author means that the average age was over 50 years, this still does not make sense for that Rascol et al 2010 and Im et al 2003 (Table 1, Line 7 and Line 21, Page 830) were included in the meta‐analysis with the data of average age unavailable.

Forth, the literature search was incomplete. A similar network meta‐analysis by Zhuo C titled “Network meta‐analysis comparison for efficacy and tolerability among ten drugs for treatment of Parkinson's disease: a network meta‐analysis”6 contained 110 included studies, while the author just included 24 studies. Additionally, some of included studies in the study by Zhuo et al was not included in the present study by Li et al, such as the study by Sethi KD,7 the study by Stern MB.8

Comprehensive evaluation of the incidences of adverse reactions produced by the various drugs is clinically useful in the selection of drugs for the treatment of Parkinson's disease. Unfortunately, in our opinion the article does not provide any valid information regarding the safety of eleven drugs used for Parkinson's disease with inaccuracy in data analysis. Hence, we expected the authors to reanalyze the data to update the current views, otherwise it will mislead more and more readers with respect to the selection of drugs for the treatment of Parkinson's disease.