4.3.1. Strengths

We used a clear research question, and a comprehensive search strategy, with methods to identify unindexed papers. We registered our study protocol with PROSPERO, followed STROBE and PRISMA guidelines when conducting our review, and independently screened and critiqued papers.

### 4.3.2. Limitations at Study Level

These have largely been covered under study quality. A major limitation for many studies was that of statistical power, which may explain why included studies provided very limited data on risk factors by gender and age-group, and why many studies aggregated successive years of data. Presenting average age and gender-specific suicide rates for periods of up to 41 years obscured any year-on-year changes and made it hard to identify short-term variations during a period of rapid social change [18]. Even where gender- and age-specific suicide rates were presented for specific years over the period 1970–1995, these lacked statistical tests for time trends, and only related to West Greenland [20] so were unlikely to be generalisable to the rest of Greenland [21]. Without join-point regression models to calculate annual percentage change statistics we lacked an understanding of the magnitude and direction of short-term and long-term trends in age-specific suicide rates.

Misclassification bias was another general potential problem. All studies derived suicide cases from population death registers, or from the death certificates and police reports on which those registers were based. Only one was questionable in terms of whether it used comprehensive methods of identifying all cases [19]. In one included study, 94% of deaths from 1968–1999 (as recorded in death certificates) had been certified by a physician and issued with an underlying International Classification of Diseases (ICD) code for cause of death [21]. This highlighted that where no diagnosis is made or where classified as other injuries, it is possible that there was under-recording of suicide, which may have been differential by age group for cultural reasons. Given high rates of accidental death in young men, and wide international variation in the quality of suicide data, the grea<sup>t</sup> potential for underestimating, through misclassification, suicide deaths in young men has been acknowledged [1]. It is possible that high rates of suicide in young men in included studies might be under-estimates. Conversely, it is possible that biases might lead to the deaths of young men being more likely to be classified as suicide than deaths in other demographic groups. Validation studies, however, sugges<sup>t</sup> that registration within official Greenlandic statistics is generally reliable [37].

### 4.3.3. Limitations at Review Level

Due to our eligibility criteria we may have overlooked studies published in Danish or Greenlandic, and our search for unpublished studies may have been incomplete, thereby reflecting publication bias. By confining our study to studies reporting suicide rates we lacked the context of age and

gender patterns in suicidal ideation and suicide attempts over a similar period, particularly where they provided gender [38] or cross-national [39] comparisons.
