DSM-IV diagnoses for all disorders were made algorithmically from SSAGA information. However for these analyses we also generated a DSM-5 diagnosis for AUD in the following way. Individual alcohol symptoms were queried, starting with symptoms of DSM-IV alcohol dependence and alcohol abuse, adding craving and subtracting legal problems related to alcohol. Onset and offset of each symptom was recorded, making it possible to cluster symptoms that occurred by age. Thus the analyses presented here use DSM-5 AUD as an outcome variable while all other disorders are diagnosed by DSM-IV.Diagnoses of externalizing and internalizing disorders at the baseline interview were also made algorithmically from the SSAGA using DSM-IV. Externalizing disorders included any of the following: ADHD, conduct disorder/antisocial personality disorder, oppositional defiant disorder, drug use disorder . Internalizing disorders included major depression, panic disorder, obsessive-compulsive disorder, social phobia, and agoraphobia.Subjects were divided into groups based on whether they had an externalizing disorder or an internalizing disorder at the time of the baseline interview. The groups were: Externalizing, Internalizing, Both, or Neither. Alcohol use disorder diagnosis was then assessed at each interview period, using the DSM 5 distinctions for Mild AUD , Moderate AUD , and Severe AUD . Subjects with age of onset of AUD prior to age of onset of internalizing/externalizing disorders were excluded from analysis. We also performed a sensitivity analysis in which all subjects with externalizing were compared with all subjects without externalizing; likewise subjects with internalizing were compared with all subjects without internalizing . An externalizing-internalizing interaction term was included in this analysis. Overall, 43.0% of the sample met criteria for a diagnosis of either Mild, Moderate, or Severe AUD by the end of the observation period .
At the time of the baseline interview, 982/3286 subjects had an externalizing diagnosis ; 140/3286 subjects had an internalizing diagnosis , 286 had both and 1878 had neither . All covariates had significant relationships to age of onset in subjects with either mild, moderate, or severe AUD . The association of any comorbid disorder and presence of Alcohol Use Disorder was significant overall ,plant growing stand and there was a significant effect of comorbidity on age of onset as well . Among subjects with an externalizing disorder only at baseline, 515/982 had some type of AUD during the follow-up period. Among subjects with an internalizing disorder only at baseline 66/140 had an AUD. Among subjects with both externalizing and internalizing, 182/286 had an AUD. In comparison, subjects with neither type of disorder had an AUD rate of 34.7% . Figure 1 shows onset of alcohol use disorders in subjects stratified by initial diagnoses of Externalizing disorder, Internalizing disorder, Both, or Neither. Figures 1a–c show onset of mild, moderate, and severe AUDs respectively. For each type of AUD, the relationship with comorbid disorders is significant by Log-rank test and Cox Proportional Hazards . Age of onset comparisons are shown in Kaplan-Meier Plots . Each of these shows significant effects of comorbidity by Log-rank Test . The plots do not include a covariate effect but we have also achieved similar results by the Cox model adjusting for covariate effects . The statistical effect of comorbidity is generally greatest in the development of Severe AUD and least in Mild AUD based on the hazard ratios in the different comorbidity types . The three groups are significantly different from each other in the strength of the comorbidity effect . The sensitivity analysis showed a clear effect of externalizing on age of onset in mild AUD, moderate AUD, and severe AUD . For internalizing, there was an effect in moderate AUD and severe AUD . No statistical interaction was seen between the effect of externalizing and the effect of internalizing. Age of onset distributions are presented for Mild AUD , Moderate AUD , and Severe AUD . The distributions include drinking milestones as well as onset ages for the diagnoses of Mild AUD , Moderate AUD and Severe AUD . As noted above, the study samples are independent of each other for analytic purposes, and are classified according to the most severe disorder that the subject met criteria for during the observation period.
Figure 2 shows drinking milestones in subjects who developed an alcohol use disorder. Figure 2a–c show mean, median, interquartile range, and outliers for subjects with mild , moderate and severe alcohol use disorder. Subjects are classified in a cohort according to the most severe form of disorder they manifested during the observation period. In Figure 2b milestones for the moderate group include the age when they would have been first classified as showing a mild AUD. In Figure 2c milestones for the severe group include the ages when they would have been first classified as showing a mild or moderate AUD. We used ANOVA and i-test to detect the correlation between the onset of drinking milestones in the four diagnostic groups. The mean age of first drink progresses from 16.2 inUnaffected to 14.9 in Mild to 14.4 in Moderate to 12.8 in Severe . The mean age of first regular drinking progresses from 18.8 in Unaffected to 17.5 in Mild to 16.9 in Moderate to 15.7 in Severe . The mean age for meeting criteria for Mild AUD progresses from 18.6 in Mild to 17.4 in Moderate to 16.1 in Severe . The mean age for meeting criteria for Moderate AUD progresses from 19.1 in Moderate to 17.3 in Severe . The age of onset for Severe AUD is 18.5. This age relationship is detailed in Figure 3. Figure 3 represents the onset of alcohol use and alcohol problems in 3286 adolescents observed over a ten year period. It includes 1870 who remained unaffected, 684 who developed mild alcohol use disorder, 415 who developed moderate alcohol use disorder, and 317 who developed severe alcohol use disorder. The ANOVA for onset of first drink among the unaffected, mild, moderate, and severe cohorts shows p < 0.001. The ANOVA for onset of regular drinking among the unaffected, mild, moderate, and severe cohorts shows p < 0.001. The ANOVA for onset age of mild AUD among the mild, moderate, and severe cohorts shows p < 0.001. The t-test for onset age of moderate AUD between the moderate and severe cohorts shows p < 0.001. These data suggest a strong effect of externalizing and internalizing disorders on prevalence and age of onset of Alcohol Use Disorder among adolescents/young adults at risk for the development of AUD on the basis of family history.
Externalizing disorders were clearly associated with an increased risk for AUD and for earlier development of AUD. Internalizing disorders by themselves did not show a significant effect, but in combination with externalizing disorders they were associated with an earlier onset for severe AUD . When we considered all internalizing disorders together a clear effect on onset of moderate AUD was seen as well. By the end of the follow-up period,plant grow table more than 60% of young people with both externalizing and internalizing disorders at baseline had developed alcohol dependence in comparison with about 30% of young people with neither type of comorbid disorder. The effect of comorbidity was stronger in more severe forms of AUD, with a 6-fold increase in risk for Severe AUD among subjects with both externalizing and internalizing disorders compared to subjects with neither form of comorbid disorder. There was also evidence for an earlier developmental course in more severe forms of AUD compared to less severe. Persons with Severe AUD were likely to have their first full drink prior to the age of 13 and be drinking regularly prior to age 16 and experiencing 1–2 alcohol problems by that same age. In contrast young people who did not demonstrate any AUD were likely to have their first drink at 16 and start regular drinking just prior to age 19. Median and mean ages of onset for each type of AUD were 18–19, though the range extended through the follow-up period. Those at greatest risk for an AUD were males of European descent from an alcohol dependent proband family with one or more childhood onset psychiatric diagnoses. Those at least risk were females of African-American ancestry from a non-case family with no childhood onset diagnosis. Limitations of the study include the fact that all analyses are based on self-report and there is no independent corroboration of diagnoses or symptoms. Subjects interviewed in their late 20s may have had more difficulty with accurate reporting of events in early teenage years in comparison to subjects in their mid-teens. Retention rate from baseline interview to two-year interview was 85%, the majority of subjects completed at least four interviews . Families in the COGA study tend to be densely affected and results may not be generalizable to persons with alcohol use disorder in the general population. The subjects were ascertained at 7 University-based clinical sites and the populations studied reflect those sites. The magnitude of these effects was substantial, and this information may be helpful in targeting efforts at education and prevention. In this sample most of the AUD-affected subjects had a comorbid psychiatric disorder at baseline. Many such subjects may come to clinical attention for their childhood-onset disorders and it may be worth educational efforts targeting AUD, especially for those at increased familial risk. It has been argued, though, that more intensive interventions are not likely to be cost- effective at this time . It seems to be of value to continue to try to quantify risk in various clinically and biologically identifiable groups. Polygenic risk scores, especially as they increase in power with data from expanding clinical samples, will likely be of use . It would also be of value to attempt to separate AUD effects from other forms of SUD, since we know that they are highly comorbid in many samples, including the sample studied here. The internet has profoundly altered how individuals obtain information regarding their health, and men’s health is no exception. Although men are less likely than women to pursue preventative health care and more likely to develop chronic cardiovascular and metabolic disease, the rise of online social media platforms may play a role in challenging this disproportion. Men contending with infertility increasingly turn to social media platforms for information, guidance, and discussion with peers. A male factor contributes to nearly 60% of all cases of infertility, yet cultural and societal constructs of masculinity create psycho social barriers to consultation with a physician. Social media platforms enfranchise men to take an active role in understanding causes and treatments for infertility by providing anonymity absent from face-to-face encounters. Although health information online is becoming more readily accessible, it escapes the scrutiny of scientific publication guidelines, allowing for the propagation of non-evidenced-based material. Social media tends to amplify the most sensational content and headlines. Literature assessing the quality of male infertility content available online remains scarce, although a recent review has shown that urological conditions as a whole suffer from a spread of misinformation on social media. Social media analytics tools have emerged that provide detailed, quantitative metrics, but these tools have not yet been applied to content in the male infertility space. Given the proliferation of sensationalism on social media, we hypothesized that content about male infertility shared on social media platforms may be largely inaccurate or misleading. Using a combination of ananalytics tool and a quality rating system, we per formed a quantitative and qualitative analysis of male infertility content shared on social media. These data may inform how men’s health specialists should aproach patient education about male infertility, as well as ways in which they engage with social media in the future.We used the social media analytics tool BuzzSumo to identify the most shared male infertility content from September 2018 through August 2019. BuzzSumo gathers data across the social media platforms Facebook, Pinterest, Reddit, and Twitter to generate a list of article links with the highest online engagement. Engagement is defined as the total number of interactions that users have with a particular article link, including actions, such as “liking,” “commenting,” and “sharing” on social media. Two urologists with advanced fellowship training in male reproductive medicine initially screened a total of 20 search terms related to male infertility using BuzzSumo.