Results

As shown in Figure 2 and supplement table 1 , comparing with the risk factor “diabetes”, most of A10 drugs demonstrated curative effects for patients with diabetes, while sglt2i might increase risk of ketoacidosis (IC025 of diabetes: 3.57 vs.IC025 of empagliflozin: 6.99), lower limb extremity amputation such as toe amputation (IC025 of diabetes: 3.38 vs. IC025 of canagliflozin: 5.35), gangrene such as Fournier’s gangrene (IC025 of diabetes: 3.75 vs. IC025 of empagliflozin: 6.76), infection such as urinary tract infection (IC025 of diabetes: 0.10 vs. IC025 of canagliflozin: 1.84), ulcer such as skin ulcer (IC025 of diabetes: 0.79vs. IC025 of canagliflozin: 2.42), peripheral ischemia (IC025 of diabetes: 0.39 vs.IC025 of canagliflozin: 2.23), kidney injury such as acute kidney injury (IC025 of diabetes: 1.56vs. IC025 of dapagliflozin: 2.27), various inflammation including osteomyelitis, fasciitis and cellulitis, especially for the case of canagliflozin. osteomyelitis (IC025 of diabetes: 1.80 vs.IC025 of canagliflozin: 4.17) and cellulitis (IC025 of diabetes: 0.56 vs.IC025 of canagliflozin: 2.16) were AEs unique to canagliflozin.
The FAERS database is composed by a total 14,073,327 AE reports from January 1 2004 to up to September 30, 2021. Following the data cleanse procedure described above, there are 2,888 osteomyelitis-related reports referring to hypoglycemic drugs (A10), among which 2,333 cases are associated with SGLT2i, mostly contributed by canagliflozin counting 2,283 (Table 1 ). Among reports referring to both canagliflozin and osteomyelitis, 73.50% of patients are male, while the gross gender ratio for each category of A10 drugs is relatively balanced. Among all osteomyelitis-related patients treated with canagliflozin, 59.22% are 30 to 49-year-old, as well as 23.74% patients aged 18 to 29-year-old, 14.76% patients aged 50 to 64-year-old, which added up 97.72% patients aged from 18 to 64-year-old, comparing to 79.07% cases are categorized in the same age group for A10 treated patients. For reports exposure to canagliflozin, 99.82% of which classed the targeted drug as primary suspect drug (PS). The most common reporting source is from consumers, counting 91.90% for canagliflozin-related case associated with osteomyelitis, on the contrary, for all reports referring to A10 drugs about 50% of are filed by consumers.
All interested drugs-osteomyelitis pair were subjected to disproportional analysis and BCPNN in duplicates with filtering for the diabetes indication, and results are demonstrated in Figure 3 . As demonstrated in Figure 3 , a total of 1,451 counts of osteomyelitis related AEs are generated out of total 1,438 reports listing canagliflozin, and the ROR value is 360.89 (95% CI 340.58-382.41) coupled with IC025 value as 7.79. Detailed to each osteomyelitis-related AE individually, canagliflozin-osteomyelitis pair counted 1,214, generated ROR as 315.60 (95% CI 296.51-335.93) and IC025 as 7.62, canagliflozin-osteomyelitis acute pair counted 157, ROR as 1391.14 (95% CI 1134.55-1705.76) and IC025 as 6.48, canagliflozin-osteomyelitis chronic pair counted 72, ROR as 716.11 (95% CI 546.95-937.60) and IC025 as 5.03. All above signals are classed as strong signals. For canagliflozin-staphylococcal osteomyelitis pair, although a high ROR value was generated as 168.49 (95% CI 81.87-346.74), but coupled with IC025 as -1.21, and therefore cast out as a negative signal by BCPNN, as false positive. For osteomyelitis associated with empagliflozin, ROR value is 2.72 (95% CI 1.22-6.06) and IC025 as -0.19 (Figure 3 ), while other SGLT2i as well as other hypoglycemic drugs (A10) except for insulin and its analogs (A10A) did not generate any valid ROR. Non-canagliflozin SGLT2i as a group could generate ROR as 1.99 (95%CI 0.95-4.17) and IC025 as -0.33, while other hypoglycemic drug groups including biguanides, DPP4, GLP1, TZD did not generate any valid ROR signal. Among all reports referring to osteomyelitis, 405 cases referring to insulin and its analogs and generated ROR as 1.32 (95% CI 1.08-1.62) and IC025 as 0.09, which could be considered as a weak signal, 484 cases referring to non-insulin hypoglycemic drugs (A10B) other than SGLT2i and could not generated valid ROR as 0.28 and IC025 as -2.25, which meant no signal. Taking gender as filter, there is significant difference in the ROR of osteomyelitis associated with canagliflozin between males (ROR 453.79, 95% CI 424.51–485.10, IC025 as 7.96) and females (ROR 190.38, 95% CI 171.07–211.87, IC025 as 6.69). For insulin-osteomyelitis pair, only male generates a weak signal (ROR 2.00, 95% CI 1.53-2.63, IC025 as 0.57).
To demonstrate the changing patter of q-ROR (Supplement Table 2& Figure 4 ), natural logarithm value of ROR (Ln ROR) was used as vertical coordinates, and plotted against quarters of year as horizontal coordinates. As shown in Figure 4 , although reporting counts of canagliflozin-osteomyelitis pair diminishes considerably with filtering diabetes as indication comparing to without doing so, the curve of canagliflozin is almost overlapped with curve of canagliflozin (wo, without). When A10 drugs excluding SGLT2i and insulin were investigated as drug group, all the q-ROR value are below the threshold of ROR value as 1 (Ln ROR as 0), while the Ln ROR-time curve of insulin (insulin and its analogs as a group, A10A) remains a generally horizontal line, with ROR value consistently within range from 1 to 2 (Ln ROR range from 0 to 1), since the second quarter (Q2) of 2013. As shown in Supplement Table 2 , during 18 years from January 1 2004 to up to September 30, 2021, q-ROR value of insulin is always above the recognition threshold of 1 and fluctuates consistently around a median as 1.57 (mean 1.71 ± 0.44), and valid ROR could be identified since Q3 of 2004. The curve of canagliflozin and SGLT2i starts to generate valid ROR signal, since as early as 2017 Q4, while for any drug group including SGLT2i, first valid ROR emerges since 2018 Q1. For any drug group excluding canagliflozin, such as non-canagliflozin A10 and non-SGLT2i A10, no valid ROR signal is ever generated (Supplement Table 2 ).
Chi2 tests were then applied to investigate correlation between the series of q-ROR, using a null hypothesis claiming the prevalence of any two-given series of q-ROR was the same. Among all series, canagliflozin, canagliflozin (wo), and SGLT2i, share the same pattern, although the scales of ROR were considerably different. Changing pattern of canagliflozin and SGLT2i demonstrates differences from insulin (p = 0.00) and other hypoglycemic drugs or drug groups (Supplement Table 3 ). Another Chi-square test was introduced to determine the prevalence of q-ROR of A10-osteomyelitis pair before and after the approval of SGLT2i during the same span of time, which is from Q2 of 2004 to Q4 of 2012 as serial A and from Q2 of 2013 to Q3 of 2021 as serial B and a p value of 0.00924 was generated, and the null hypothesis was rejected.