Discussion

As shown in Figure 2 , ketoacidosis, various infection, peripheral ischemia, renal impairment, inflammation including osteomyelitis might more likely to occur among SGLT2i users, especially canagliflozin. Our findings suggested that SGLT2i increased the risk of these issues, or was less effective on them. SGLT2i treatment for patients who suffered from ketoacidosis, cardiovascular issues, renal problem and inflammation was therefore not recommended. Osteomyelitis and cellulitis are AEs unique to canagliflozin. Since osteomyelitis is considered to greatly increases the risk of a lower-extremity amputation [17, 18], our findings indicated that exposure to canagliflozin could notably increase the risk of developing osteomyelitis while other A10 drugs could reduce such a risk. On the other hand, these events could be monitored as warning prior to lower limb extreme amputation.
Most osteomyelitis-related cases were referred to canagliflozin, indicated that there might be a strong correlation between SGLT2i exposure, especially canagliflozin, and developing osteomyelitis according to FAERS data. In this study, ROR and BCPNN method were applied to investigate association between hypoglycemic drugs and osteomyelitis. Signals with high value of ROR were detected for canagliflozin-osteomyelitis pairs, as well as any drug group containing canagliflozin. Since the value of ROR did not directly indicate the significance of a signal, all positive signals were validated by BCPNN method, strong signals were generated canagliflozin or any drug groups containing canagliflozin associated with osteomyelitis, while weak signal was generated for insulin (A10A-osteomyelitis pair), and no signal was ever generated for other A10 drugs or drug groups excluding canagliflozin as well as insulin. Therefore, these findings indicated that association between canagliflozin treatment and osteomyelitis was quite convincing. Weak signal generated by insulin-osteomyelitis pair, might be explained by the insulin exposure as well as morbidity of diabetes since insulin treatment indicating a proxy of disease severity or disease advanced stage [26, 27], but the morbidity of diabetes might neither be a sufficient condition nor necessary condition for a patient with diabetes to develop osteomyelitis. Case number of targeted drugs presented notable differences with or without filtering diabetes as indication, and IC025 of canagliflozin-osteomyelitis pairs with the filtering was lower than without it, which could lead to the conclusion that casting out cases without a specific indication as diabetes resulted in dwindling the intensity of BCPNN signal.
Among patients developed osteomyelitis, 73.50% of whom are male, while the gross gender ratio for each category of drugs is relatively more balanced (Table 1 ). Further disproportional analysis using gender as filtering was applied and results suggested that there was significant difference in the ROR of canagliflozin-osteomyelitis pairs between two genders. Since the cleansing procedure according to including/excluding criteria mention above had excluded all reports with infection know as competing indication and reaction [29], the gender ratio as well as differences of ROR and IC025 value between male and females was possibly due to sex differences, negative correlation might existed between glycosylated hemoglobin (HbA1c) and serum testosterone levels [31]. As displayed by insulin-osteomyelitis pair, only the dataset of male patients could generate a valid ROR and a weak signal of BCPNN, these findings support the hypothesis that male patient might more likely to develop osteomyelitis. When exposure to SGT2i, and canagliflozin were the major factor for causing disproportionality, dataset of male could generate ROR value three times larger than female. When the signals were validated by BCPNN method, both genders generated strong signals (IC025 7.96 for male vs. IC025 6.69 for female), which indicated that for reports of each gender, however their difference in ROR, canagliflozin presented a strong correlation with developing osteomyelitis.
A new approach so-called quarterly ROR (q-ROR) was introduced to demonstrate the ROR developing pattern, and the series of q-ROR generated by different drugs or drug groups was subjected to Chi-square tests to determine their correlations statistically. Canagliflozin (wo), canagliflozin, SGLT2i and A10 demonstrates no prevalence difference although there might be gap in the scale of ROR value (Supplement Table 2 & Figure 4 ) and supported the speculation mentioned above that the disproportionality of osteomyelitis-related reports was generated by canagliflozin. For hypoglycemic drugs other than canagliflozin and drug groups excluding SGLT2i, no positive signal was ever generated. These findings strongly indicated that developing pattern of these drugs or drug groups were synchronised by the presents canagliflozin.
In pharmacovigilance study, disproportionality emerges when a specific adverse event is associated with a given drug [32-34]. In this study, we tried this approach called q-ROR to utilize the FAERS data quarterly, and mimicked accumulation of reports to the database in real world. Starting from a setting date, slice of data was added to dataset in chronologically order on quarterly basis and a ROR value from the setting date up to that quarter was calculated. A serial of RORs was generated for any given interested drug (drug group)-AE pair. Finally, value of q-ROR achieved equilibrium approached its theoretically true value. For those lately approved drugs with limited reports but analogs of which had been long approved, the q-ROR curve might be used to predict its association with interested AE according to their precursors or as a drug group, such as ertugliflozin, luseogliflozin, remogliflozin and other new SGLT2i fail to generate any positive signal. The q-ROR value of the insulin-osteomyelitis is always above the recognition threshold and fluctuate consistently around 1.5 and positive signal could be identified since Q1 of 2005, while series of q-ROR referring to any drug group excluding canagliflozin and insulin went beneath the thread hold of 1, since 2005. Coupled with the dramatic increasing reports of canagliflozin related osteomyelitis in FAERS (25 cases in 2017 vs. 1,402 cases in 2018) (Table 1 ), the q-ROR pattern of canagliflozin bounds from 3.24 in Q4 of 2017 to 79.54 in Q4 of 2018 (Supplement Table 2 ).This finding indicated that, q-ROR could be used to monitor drug-induced ADRs unknown to premarketing trials as pharmacovigilance, when a dramatic rise in ROR value was spotted for given drug-AE pairs, and needed to be further vilified by BCPNN method.
Diabetes was a competing risk factor of developing osteomyelitis, which occurs in approximately 10-20% of patients with diabetes-related foot ulcers [17], and osteomyelitis of the lower extremity is a commonly encountered problem in patients with diabetes [35]. In this study, such a data pack in FAERS was equivalent to considering all hypoglycemic drugs as a drug group and filtering with diabetes as indication which generated a signal considered to be caused by both the treatment and the morbidity. And this data pack was also examined by quarterly ROR method (q-ROR). Chi2 test was induced to comparing q-ROR value prior to and since canagliflozin was approved in Q1 of 2013, and a p value of 0.00924 < 0.05 indicated that ROR developing pattern experienced a significant change, which probably due to the exposure to SGLT2i, especially canagliflozin, since before the approval of SGLT2i, ROR serial of drug-event pair generated no positive signal and the indication was confined as diabetes mellitus.
Although previous publication suggested that osteomyelitis of the lower extremity is a commonly encountered problem in patients with diabetes [35] and occurred in approximately 10-20% of patients with diabetes-related foot ulcers [17]. According to our findings, q-ROR series of drug groups excluding canagliflozin and insulin generates no positive signal (Supplement Table 2 & Figure 4 ). q-ROR serial of insulin presented all its value of q-ROR ranging 1.71 ± 0.44, with median as 1.57, as well as ranging 1.53 ± 0.25, with median as 1.45, since Q2 of 2013, when canagliflozin was approved as the first of SGLT2i, which indicated that the morbidity of diabetes mellitus, even at a proxy of disease severity or disease advanced stage [26, 27] might not be consider as a significant interfering factor for drug-associated osteomyelitis based on FAERS, otherwise all drug groups filtering diabetes as indication should consequently generate ROR signals above 1. And therefore, it strengthened the results of Chi2 test between q-ROR series, that was, canagliflozin exposure might be the predominant cause of developing osteomyelitis for patients with or without filtering diabetes as indication, based on FAERS database. On the contrary, other widely used SGLT2i, such as dapagliflozin and empagliflozin might not associated with developing osteomyelitis, and for those lately approval SGLT2i which had not accumulated enough ADR reports for disproportional analysis yet, predictions could be made base on the q-ROR pattern as pharmacovigilance.
There are certain limitations might undermine this study. Spontaneous reporting systems including FEARS were exposed to the biases inherent to pharmacovigilance studies. To the best of our knowledge, in 2018, Chang, H.Y., et al. [11] had mentioned about risk of osteomyelitis when discussing about the association between SGLT2i treatment and lower extremity amputation among patients with T2D, and osteomyelitis of the lower extremity is a commonly encountered problem in patients with diabetes [35] and occurs in approximately 10-20% of patients with diabetes-related foot ulcers [17]. These publications coincidently match with a gush of osteomyelitis related ADR reports and a surge in ROR for canagliflozin-osteomyelitis pair.