TABLE 2.

Quality assessment, sample selection, measurement scales, analysis, and interpretation of findings.

Identifiers and quality assessment			Sample selection		Measurements and analysis		Results and discussion
References	Number	Grade	Selection of participants	Retaining participants/ Compensation	Measurements	Statistical analysis	Interpretation	Discussion
Herawati et al. (10)	1	Medium	One exclusion criterion was age > 30 years old, however, the table of characteristics of study subjects reported range of 17–49 years. The Snowball Sampling technique is inappropriate, since college student should be easily contactable by their school, therefore voiding the justification of them being “hard to reach population”	Not applicable	The use of Hamilton Anxiety Rating Scale is widely accepted, and the authors mentioned that several local studies have used them.	The use of logistic regression is acceptable, given the binary outcome of interest	The authors seem to interpreted Odds Ratio as Risk, which is problematic. The “event” is not rare, therefore the OR will not estimate RR	Supported by results, however, the comparison to China is quite far-fetched, since the majors are different, and the tools used to measure anxiety are not HARS
Ugidos et al. (40)	2	High	Recruitments were based on existing database of students and workers, therefore the elderly are less represented. But this was discussed and acknowledged by the authors (i.e., due to impracticalities in recruiting elderly using offline forms)	Significant loss of participants at subsequent time points, due to “loss of interest.” The authors did not seem to foresaw this and made some efforts to maintain retention (i.e., through incentives or supports)	The questionnaires used were tailored to match the objectives, which is appropriate → the InDI-D used “Presence of COVID-19” as the condition	The use of Linear Mixed Model is acceptable. The authors did not perform imputation due to the missingness properties not known (quite possible MNAR, Missing Not At Random)	In line with the reported models	It seems that the questionnaires used provide acceptable fit, the Pseudo-R squared (Tables 2, 3) are quite good, given the number of items and the three time points. These supported the interpretation of the models.
Shokri et al. (41)	3	Medium	There seems to be unclear procedures on participants recruitment. “the questionnaire was shared to participants through email, Instagram direct, WhatsApp direct, and Telegram groups.” Did they stop as soon as the recruited participants reached 1,000? Were the methods carried out in parallel?	Not applicable	The Berger HIV stigma scale, modified for Ebola was used. However, the concept of internalized stigma of Berger’s and COVID-19 is very distinct. This implication was not discussed.	The use of ANOVA and T-test are acceptable. The significance level was set at 0.005, but subsequently at 0.05 in the results (most likely a typo in the method section)	Straight-forward and concise interpretation.	Supported by the data, particularly the context of governmental jobs and self-employed situation were discussed.
Amir (16)	4	Low	It is highly disturbing that “saturation principle” was used to stop the recruitments, but this was based on a single interpreter/coder, without cross examinations. Thus whether saturation was truly achieved is questionable.	Not applicable	Seven-phase data analysis framework was used but all identifications of themes relied on this single author → very prone to coding bias	Pure qualitative study, without statistical analysis	All of the themes are supported by actual quotes by the participants	Supported by the interviews
Miconi et al. (17)	5	Medium	Based on Leger Opinion Panel. Therefore some degree of selection bias is expected (non-participants of Panel would not be able to participate)	The authors’ decision to reward participants based on survey completion time was highly questionable (CAD 0.5–2). This practice has been discouraged by survey guidelines.	The use of Hopkins Symptom Checklist-10 is acceptable, however, to my knowledge it has not been validated in Arab population (which contributes n = 450 in this study)	The use of factor analysis to dimensionally reduce the data complexities of HSCL-10 is acceptable	Supported by the results	The discussion draws comparison to the UK and US which investigated COVID-19 in multi-racial settings. The recommendations (focusing on ethnic minorities) are supported by the findings in the UK and US
Yuan et al. (12)	6	High	COVID-19 survivors were compared with healthy control of the same city (by convenience sampling).	Not applicable	Fatigue and Stigma were assessed using PHQ-9 and Social Impact Scale (a generic stigma scale). The SIS has been validated in Chinese population previously, but not for COVID-19	T-test, ANOVA, and generalized linear model were used, which are appropriate.	The interpretation were based on the comparison between survivors and healthy controls, appropriate given the study design’s limitation of convenience sampling.	The recommendations were based on the findings, however, the decision to not performed matched controls was not explained. Almost half of the survivors were male, but only 20.6% were male in the control group. This imbalance might skew the findings toward the null
Lohiniva et al. (8)	7	High	Based on maximum variation and data saturation principles, which are standard practices for family-based interviews	Not applicable	Framework analysis following the Health Stigma Framework. Four investigators discussed the coding and framework analysis	Pure qualitative study, without statistical analysis	All of the themes are supported by actual quotes by the participants. Attributions of the quotes were presented consistently	The themes and recommendation being discussed were based on the interview results
Tomczyk et al. (11)	8	High	Based on online advertisements through Facebook.	Participants received EUR 5 if they finished the set of questions	Mostly based on mental health questionnaires. But some of the adaptation was quite questionable, for example the “Persons with COVID-19 should not be allowed to have a driver’s license” question	Since the classes were 3, producing three pairs of 1-2, 2-3, and 1-3, the use of Multinomial Logistic Regression is appropriate.	In line with the reported results	The tailored health promotion efforts targeting youths was supported by the RRR results
Jiang et al. (4)	9	High	The study has a specific objective of exploring stigma of Chinese people toward people from Wuhan, therefore the recruitment method is acceptable	Not applicable	Adaptation of stigma related to tuberculosis was used. Pre-testing was performed to ensure appropriateness to COVID-19	Chi-square and logistic regression were used, which are appropriate.	In line with the reported results	There are several inconsistencies in the discussion section. The authors acknowledged that the sampling was not representative and not probabilistic, but they presented and discussed the findings as prevalence. This is clearly inappropriate.
Mahmoudi et al. (19)	10	Medium	Convenience sampling method is deemed acceptable, since they objective of the study was to have PCR test and Chest CT. One minor concern is why the four hospitals were used instead of the others was not properly explained.	Not applicable	PTSD and Self-Stigma Scale-Short (originally was designed for mental health, immigrant, and sexual-orientation minority groups) were used. Adaptation for COVID-19 was performed. Insomnia and MHI-5 were also used	Since the classes were 3, producing three pairs of 1-2, 2-3, and 1-3, the use of Multinomial Logistic Regression is appropriate.	The authors seem to be a bit over optimistic in reporting the goodness of fit. The root mean square error of approximation (RMSEA) value indicated close to mediocre fit, not “satisfactory” as the authors have implied	The mediation analysis were quite acceptable, however, the discussion did not acknowledge that the fit was rather mediocre → some unexplored and unmeasured factors might be necessary to be investigated to improve the current mediation analysis
Earnshaw et al. (9)	11	Low	Based on Amazon MTurk Panel. Therefore some degree of selection bias is expected (non-participants of MTurk would not be able to participate)	Participants received USD 2 if they finished the set of questions	Chronic Illness Anticipated Stigma Scale was used, but only relevant items for COVID-19 were retained	Linear regression was used for the analysis	The decision to display testing likelihood without confounder adjustment is questionable. Gender and Race (two covariates which were collected in the study) should be included in the analysis.	The interpretation is greatly limited by potential confounders not considered by the authors.
Kang et al. (18)	12	High	Retrospective analysis of medical record of a community isolation facility	All of the patients analyzed. Time cut-off were used to exclude participants.	The use of MERS questionnaire was appropriate, given the relative similarity of transmission methods between MERS and COVID-19 viruses.	Acceptable. Limited number of participants complicate further analysis of the data. For example, only n = 32 have been isolated up to the 4th week. Therefore the absence of further analysis is understandable	In line with the reported results	The authors promoted the CTC model in handling COVID. Notably the comparison to other models were given with careful consideration that they are in other countries with different system and different diseases (H1N1)
Singh et al. (13)	13	High	Participants were cohort participants of CARRS and India-UDAY. Re-contacted to participate in the current research.	No information was given on the approach to recruit participants, and whether refusal translate to exclusion from cohorts future investigations (which may dis-incentivize participants)	The anxiety was assessed by generalized anxiety disorder questionnaire, an established method. The remaining items are specifically developed by the authors for this study (without adequate validation or pre-testing)	For the qualitative part the authors used “illustrative non-attributable quotations,” without any justification. The data collection protocol recruited a diverse set of participants (i.e., not a very specific group of people), therefore attribution should be performed.	Correct interpretation of the Odds Ratios and factors that changes the OR (with specific examples for ease of interpretation)	The tailored health promotion efforts targeting youths was supported by the RRR results
Bodrud-Doza et al. (14)	14	Medium	Email and social platform recruitments (Facebook, WhatsApp, etc.) combined with targeted database of hard to reach group.	Not applicable	The questions were specifically developed for COVID-19 instead of adapting them from previous questionnaires. Expert consultation was used to validate the questions	There were 46 items which the authors attempted to be reduced to a manageable levels of number of variables for easier interpretation. The use of CTT PCA sequentially followed by CA were unusual, but not unheard of.	There is one major concern regarding the Scree Plot and Eigen-cutoff. The alternative of PC1-PC5 seems to be more appropriate, instead of including nine PCs (Figure 2 and Table 3)	The authors provide additional information and suggestion which were reasonable, but not supported by the data, particularly in the “Disadvantaged communities” section.
Zakar et al. (15)	15	High	A mix of purposive and snowball sampling. This is a bit unusual, but justifiable since the objective was to capture participants from multiple cities.	Not applicable	Theme identifications were performed by multiple analysts, and difference in coding/interpretation were resolved through discussion → considered to follow the recommended practice.	Pure qualitative study, without statistical analysis	All of the themes are supported by actual quotes by the participants. There are inconsistencies in quote attribution, however. For example, several quotes only had “one participant said” but some have age, gender and occupation. This reduces the trustworthiness of the quotes.	The themes and recommendation being discussed were based on the interview results