CODE protocol

Information on cause of death was recorded by 16 cohorts (Appendix) either as International Classification of Diseases, Ninth Revision (ICD-9) or Tenth Revision (ICD-10), Classification based on the Coding of Death in HIV (CoDe) project (available at: http://www.chip.dk/CoDe/tabid/55/Default.aspx), or free text. We adapted the CoDe protocol [14 ] to classify causes of death into mutually exclusive categories. Clinicians classified deaths using summary tables of patient details that included ICD-9/ICD-10 codes or free text for cause of death, patient characteristics at ART initiation (age, sex, transmission risk group, AIDS-defining conditions, and hepatitis C status), time from ART initiation to death, AIDS-defining conditions after starting ART, latest CD4 (within 6 months of death), and whether a patient was on ART at time of death. A computer algorithm developed by the Mortalité 2000–2005 Study Group [15 (link)] classified deaths using ICD-10 codes, when available. When ICD-10 codes were not available, 2 clinicians independently classified each death. Disagreements between clinicians and/or computer-assigned codes were resolved via panel discussion as per the CoDe protocol described previously [9 (link)]. Further information on rules to classify deaths is provided in the Supplementary Data.
We grouped causes of death with a frequency of <20 as “other.” AIDS was defined according to the 1993 Centers for Disease Control and Prevention classification [16 ] and included CoDe 01 (unspecified AIDS), 01.1 (AIDS infection), and 01.2 (AIDS malignancy). We grouped deaths as AIDS, non-AIDS infection, liver-related (hepatitis and liver failure, CoDe 03 and 14), non-AIDS nonhepatitis malignancy (CoDe 04), cardiovascular disease (myocardial infarction/ischemic heart disease, CoDe 08, and stroke, CoDe 09), heart/vascular disease (heart failure/unspecified, and other heart disease, CoDe 24), respiratory disease (chronic obstructive lung disease, CoDe 13 and 25), renal failure (CoDe 15), disease of the central nervous system (CoDe 23), unnatural deaths (accident/violent, suicide, and overdose, CoDe 16, 17, and 19), and other (CoDe 90).

The empirical data of this manuscript are predominantly derived from previous studies [3] (link),[15] (link),[27] (link). The modeling and results presented here were implemented in Matlab 7.6 (Mathworks, MA). We provide the code in the Protocol S1 of Supplementary Materials, enabling concrete and unambiguous specification of the computing methods employed, and the possibility to further explore the parameter space. This computation was chosen to closely mimic procedures from empirical work [3] (link),[15] (link),[27] (link).
To test local and meridional anisotropy a finely meshed grid in the visual field was projected through the models. Squares of the grid were oriented in such a way that one side was orthogonal to eccentricity, while the other side was orthogonal to polar direction. In principle, anisotropies can be derived analytically [15] (link), however the computational approach implemented for this manuscript allows flexible and comparable testing of model variations. Since we provide the code, the reader can easily implement alternative model functions within the code and test these using the methods provided.
Local anisotropy for a given position in the projection was then calculated as the length ratio of the side oriented parallel to isoeccentricity lines (i.e. M_e) divided by the length of the side parallel to isopolar lines (M_p). Meridional anisotropy is calculated based on the surface area of a set of squares with the same eccentricity, but varying polar position. Meridional anisotropy for a given position in the projection was then calculated as the surface of a square at this position (M_a(P,E′)) divided by the surface of a square at the horizontal meridian in V1 (M_a(0,E′)). Predicted areal magnification M (Figure 6c, Figure 9c) was estimated by projecting isoeccentricity bands. Areal magnification is then the square root of the projected surface divided by the surface in visual space. Analytical considerations [15] (link), have shown that this estimate of M is the most informative.

To increase the accuracy of coding and data entry, each article was initially coded by two raters. Subsequently, the same article was independently coded by two additional raters. Coders extracted several objectively verifiable characteristics of the studies: (a) the number of participants and their composition by age, gender, marital status, distress level, health status, and pre-existing health conditions (if any), as well as the percentage of smokers and percentage of physically active individuals, and, of course, the cause of mortality; (b) the length of follow up; (c) the research design; and (d) the aspect of social relationships evaluated.
Data within studies were often reported in terms of odds ratios (ORs), the likelihood of mortality across distinct levels of social relationships. Because OR values cannot be meaningfully aggregated, all effect sizes reported within studies were transformed to the natural log OR (lnOR) for analyses and then transformed back to OR for interpretation. When effect size data were reported in any metric other than OR or lnOR, we transformed those values using statistical software programs and macros (e.g., Comprehensive Meta-Analysis [24] ). In some cases when direct statistical transformation proved impossible, we calculated the corresponding effect sizes from frequency data in matrices of mortality status by social relationship status. When frequency data were not reported, we recovered the cell probabilities from the reported ratio and marginal probabilities. When survival analyses (i.e., hazard ratios) were reported, we calculated the effect size from the associated level of statistical significance, often derived from 95% confidence intervals (CIs). Across all studies we assigned OR values less than 1.00 to data indicative of increased mortality and OR values greater than 1.00 to data indicative of decreased mortality for individuals with relatively higher levels of social relationships.
When multiple effect sizes were reported within a study at the same point in time (e.g., across different measures of social relationships), we averaged the several values (weighted by standard error) to avoid violating the assumption of independent samples. In such cases, the aggregate standard error value for the lnOR were estimated on the basis of the total frequency data without adjustment for possible correlation among the averaged values. Although this method was imprecise, the manuscripts included in the meta-analysis did not report the information necessary to make the statistical adjustments, and we decided not to impute values given the wide range possible. In analyzing the data we used the shifting units of analysis approach [25] which minimizes the threat of nonindependence in the data while at the same time allowing more detailed follow-up analyses to be conducted (i.e., examination of effect size heterogeneity).
When multiple reports contained data from the same participants (publications of the same database), we selected the report containing the whole sample and eliminated reports of subsamples. When multiple reports contained the same whole sample, we selected the one with the longest follow-up duration. When multiple reports with the same whole sample were of the same duration, we selected the one reporting the greatest number of measures of social relationships.
In cases where multiple effect sizes were reported across different levels of social relationships (i.e., high versus medium, medium versus low), we extracted the value with the greatest contrast (i.e., high versus low). When a study contained multiple effect sizes across time, we extracted the data from the longest follow-up period. If a study used statistical controls in calculating an effect size, we extracted the data from the model utilizing the fewest statistical controls so as to remain as consistent as possible across studies (and we recorded the type and number of covariates used within each study to run post hoc comparative analyses). We coded the research design used rather than estimate risk of individual study bias. The coding protocol is available from the authors.
The majority of information obtained from the studies was extracted verbatim from the reports. As a result, the inter-rater agreement was quite high for categorical variables (mean Cohen's kappa = 0.73, SD = 0.13) and for continuous variables (mean intraclass correlation [26] (link) = 0.80, SD = .14). Discrepancies across coding pairs were resolved through further scrutiny of the manuscript until consensus was obtained.
Aggregate effect sizes were calculated using random effects models following confirmation of heterogeneity. A random effects approach produces results that generalize beyond the sample of studies actually reviewed [27] . The assumptions made in this meta-analysis clearly warrant this method: The belief that certain variables serve as moderators of the observed association between social relationships and mortality implies that the studies reviewed will estimate different population effect sizes. Random effects models take such between-studies variation into account, whereas fixed effects models do not [28] . In each analysis conducted, we examined the remaining variance to confirm that random effects models were appropriate.

Our analysis of the interviews drew on these field notes together with further reference to the interview transcripts. We used an interpretive inductive thematic approach, which involves identifying patterns across the interview responses rather than seeking to test pre-established hypotheses. This post-positivist approach to social inquiry is directed at identifying “making the mundane, taken-for-granted, everyday world visible” through interpretative and narrative practices (35 (link)). As sociologists, we were interested in identifying the logics that people drew on when explaining their experiences, the social relationships, connections and practices in which they took part and how they described their emotional responses (what it felt like to live during this time). This is an “analytically open” approach which attempts to explore the multi-faceted dimensions of everyday life (36 (link)). Our approach therefore did not follow a standard “coding protocol,” as we do not view the process as a linear “coding” process. Instead, our analytical process was as follows. Both authors independently began their analyses, identifying themes and cutting and pasting relevant excerpts from the interview transcripts under these themes. The authors then iteratively collaborated in deciding on which themes to highlight and in writing the analysis presented here, passing versions of our analyses back and forth and refining and editing each other's work as we did so. We acknowledge that any researcher comes to analysis with their own perspective and that analytical collaboration is a process of sharing and mulling over other collaborators' interpretations as we reach consensus over how to present our findings. Adopting this reflexive approach means eschewing a positivist perspective on qualitative research in which proof of “rigour,” “objectivity” and “reliability” is sought. Instead, researcher subjectivity is treated as resource for interpretation of research materials that can always only ever be partial and contextually situated (37 (link)).

One hundred eighty students participated in this study, including ninety healthy males (mean ± SD: age: 15.11 ± 4.79 years, height: 161.4 ± 17.13 cm, body mass: 53.14 ± 17.79 kg) and ninety healthy females (mean ± SD: age: 15 ± 5 years, height: 155.31 ± 12.28 cm, body mass: 49.22 ± 14.11 kg). Participants were from different age groups, including before puberty (i.e., 9 to 10 years; n = 30 male and 30 females), puberty (i.e., 14 to 15 years; n = 30 males and 30 females) and adulthood (20 to 22 years; n = 30 male and 30 females). A general health questionnaire was completed by participants and did not present any medical restrictions. Individuals aged 9–10 years old, 14–15 years old, and 20–22 years old participated in two hours of physical education per week at school or university. All participants who engaged in other physical activity or sports were eliminated from this study. Puberty was assessed and verified according to the Tanner model [28 (link)], which was used for inclusion or exclusion between groups. After a detailed presentation of the investigation's objectives, advantages, and potential risks, all participants and their parents provided written informed consent. The study was conducted according to the Declaration of Helsinki for human experimentation. The local research ethics committee of the High Institute of Sport and Physical Education of Kef, University of Jendouba, approved the protocol with the code number a10-2019, authorized on January 25, 2019.