3. Risk of bias in included observational studies.

Study	Representative study group	Complete follow‐up assessment	Blinded outcome assessor	Adjustment for important confounders	Well‐defined study group	Well‐defined follow‐up	Well‐defined outcome	Well‐defined risk estimation
Cunha 2004	No, based on additional information provided by trial authors, the study group described did not consist of more than 90% of the original cohort and was not a random sample.	Yes, outcome was assessed for 60% to 90% of the study group at the end date of the study.	Unclear whether outcome assessor was blinded to glucocorticoid treatment		Yes, treatment protocol and (cumulative) dose, type, duration, and form of cessation of glucocorticoid treatment were mentioned.	Yes, length of follow‐up and frequency of measuring were mentioned.	Yes, methods of detection were described, and outcome definition was objective and precise.
Felner 2000	Yes, based on additional information provided by trial authors, the study group described consisted of more than 90% of the original cohort.	Yes, outcome was assessed for 60% to 90% of the study group at the end date of the study.	Unclear whether outcome assessor was blinded to glucocorticoid treatment		No, treatment protocol was not mentioned.	Yes, length of follow‐up and frequency of measuring were mentioned.	Yes, methods of detection were described, and outcome definition was objective and precise.
Kuperman 2001	Yes, based on additional information provided by trial authors, the study group described consisted of more than 90% of the original cohort.	Yes, outcome was assessed for 60% to 90% of the study group at the end date of the study.	Unclear whether outcome assessor was blinded to glucocorticoid treatment		No, treatment protocol was not mentioned.	Yes, length of follow‐up and frequency of measuring were mentioned.	Yes, methods of detection were described, and outcome definition was objective and precise.
Mahachoklertwattana 2004	Unclear whether the study group consisted of more than 90% of the original cohort, or if it was a random sample	Yes, outcome was assessed for 60% to 90% of the study group at the end date of the study.	Unclear whether outcome assessor was blinded to glucocorticoid treatment		Yes, treatment protocol and (cumulative) dose, type, duration, and form of cessation of glucocorticoid treatment were mentioned.	Yes, length of follow‐up and frequency of measuring were mentioned.	Yes, methods of detection were described, and outcome definition was objective and precise.
Perdomo‐Ramírez 2016	Unclear whether the study group consisted of more than 90% of the original cohort, or if it was a random sample	Yes, outcome was assessed for 60% to 90% of the study group at the end date of the study.	No, outcome assessor was not blinded to glucocorticoid treatment. This information was based on additional information provided by trial authors.		Yes, treatment protocol and (cumulative) dose, type, and duration of glucocorticoid therapy were mentioned.	Yes, length of follow‐up and frequency of measuring were mentioned.	Yes, methods of detection were described, and outcome definition was objective and precise.	Yes, mean difference was calculated.
Petersen 2003	Yes, the study group described consisted of more than 90% of the original cohort.	Yes, outcome was assessed for 60% to 90% of the study group at the end date of the study.	Unclear whether outcome assessor was blinded to glucocorticoid treatment.	Yes, important prognostic factors or follow‐up was taken into account.	Yes, treatment protocol and (cumulative) dose, type, and duration of glucocorticoid treatment were mentioned. Information on the method of cessation of glucocorticoid treatment was based on additional information provided by trial authors.	Yes, length of follow‐up and frequency of measuring were mentioned.	Yes, methods of detection were described, and outcome definition was objective and precise.	No, risk ratio, odds ratio, attributable risk, linear or logistic regression model, mean difference, or Chi2 statistic was not calculated.
Rix 2005	Yes, the study group described consisted of more than 90% of the original cohort.	Yes, outcome was assessed for 60% to 90% of the study group at the end date of the study.	No, outcome assessor was not blinded to glucocorticoid treatment.		Yes, treatment protocol and (cumulative) dose, type, and duration of glucocorticoid treatment were mentioned. Information on the method of cessation of glucocorticoid treatment was based on additional information provided by trial authors.	Yes, length of follow‐up and frequency of measuring were mentioned.	Yes, methods of detection were described, and outcome definition was objective and precise.
Salem 2015	Unclear whether the study group consisted of more than 90% of the original cohort, or if it was a random sample	Unclear whether outcome was assessed for 60% to 90% at the end date of the study.	Unclear whether outcome assessor was blinded to glucocorticoid treatment	Yes, important prognostic factors or follow‐up was taken into account.	No, duration of tapering of glucocorticoid treatment was not mentioned.	Yes, length of follow‐up and frequency of measuring were mentioned.	Yes, methods of detection were described, and outcome definition was objective and precise.	Yes, mean difference was calculated.