Glucose patterns following alcohol and illicit drug use in young adults with type 1 diabetes: A flash glucose monitoring study

Abstract Introduction To assess the effects of alcohol and illicit drug use in young adults (age 18–35) with type 1 diabetes (T1D) on flash glucose monitor sensor glucose (SG) readings. Methods Twenty young adults with T1D were enrolled from a tertiary referral hospital outpatient department in Melbourne, Australia for a 6‐week prospective observational study using flash glucose monitoring (FGM). Glucometrics comparing substance using days (SUEDs) to those without substance use (non‐SUEDS) were analysed. The primary outcomes were the difference in mean SG values, its standard deviation and minutes/24‐h period out of range (SG <3.9 mmol/L or >10.0 mmol/L) between matched SUEDs vs non‐SUEDs. An interaction model with the primary effect of HbA1c on SG values was also performed. Results There were no differences in the primary outcome measures between SUEDS and non‐SUEDs. However, there were differences in the regression coefficients for HbA1c and glucometrics between non‐SUEDs and SUEDs for mean SG, time out of range and time with SG > 10 mmol/L. This difference was also identified between non‐SUEDS and days of ≥40 g alcohol for mean SG. Conclusions While there was no difference between glucometrics for SUEDs and non‐SUEDs on primary outcomes, HbA1C was found to be a less reliable predictor of glucose patterns in the 24‐h period following substance use than control days. Young adults with T1D need to monitor and respond to their glucose levels following substance use and engage in harm minimisation practices irrespective of baseline glucose control.

such technology is flash glucose monitoring (FGM), which involves the use of a factory-calibrated sensor inserted into subcutaneous tissue. The sensor measures prevailing interstitial glucose levels at frequent intervals and the results are displayed on a reader which scans the sensor. FGM provides a 24-h glucose profile and reduces the need for finger pricks. 16 Randomized trials and observational studies have confirmed the acceptability of FGM in adults with T1D, as well as its effectiveness in optimising glucose parameters such as reducing time in hypoglycaemia. [16][17][18] To date, FGM has not been used as a research tool to document glucose levels following the use of alcohol and illicit drugs. The aim of this study was to use FGM as a novel method of documenting the interaction between substance consumption and metabolic control and to establish the specific effects of substance use on glucose metrics in a real-world environment in young adults with T1D.

| Study design and participant selection
Participants in this six-week prospective observational study were recruited from an outpatient adult T1D clinic at St Vincent's Hospital, Melbourne. They were aged 18-35 years and had been living with T1D for >1 year. Participants were identified by a diabetes educator and consented for the study by the first author. Inclusion criteria were consumption of ≥40 g of alcohol (typically three to four 375 ml bottles of beer or half a bottle of wine) on a single occasion and/or had taken an illicit drug in the previous six weeks with stated intention to do so again in the coming 6 weeks. Exclusion criteria included high risk substance use, defined as a World Health Organisation ASSIST 19 score ≥ 27, or treatment for an alcohol or drug use disorder in the previous 6 months. Also excluded were people with high baseline levels of psychological distress, defined as a Kessler Psychological Distress Scale (K10) score of >20 20 or a known diagnosis of schizophrenia or psychotic disorder not related to drug use. Participants with an admission for diabetic ketoacidosis within the previous month or a Gold hypoglycaemia score of ≥4 were also excluded. 21

| Procedures
Participants were enrolled between April 2017 and January 2019.
At study enrolment, demographics, medical history, drug and alcohol history, physical examination and an HbA1c from within three months of recruitment were recorded, along with regular medications and scores on Gold hypoglycaemia, K10, WHO ASSIST and Problem Areas in Diabetes (PAID) questionnaires. 22 Females performed a urine βHCG and agreed to maintain effective contraception during the trial.
Upon enrolment, participants were supplied with a FGM reader and three glucose sensors (Abbott Freestyle ® Libre). As each sensor lasts 14 days, this provided six weeks of FGM data. The sensors were applied and used as per the product license in Australia with assistance from a physician or diabetes educator. Participants were asked to keep an activity diary to record daily alcohol and drug use.
Any changes to regular routine, such as sick days, changes in diet, travel or vigorous exercise were recorded. It was also noted whether the day was a work or weekend day. Participants were reviewed in person fortnightly for collection of activity diaries, SG data and information regarding adverse events. Participants continued regular diabetes management and were provided with contact numbers for support.

| Matching
Each participant diary was reviewed and substance using days (SUED) were identified as those involving alcohol, cannabis, stimulants (cocaine, ecstasy, methamphetamine or other) or polysubstance (three or more illicit drugs). The amount of alcohol consumed was also recorded. Each SUED was assigned a matched nonsubstance-using (non-SUED) day by two researchers blinded to the SG data. The matching day was assigned by reviewing the activity diary of each participant and selecting a comparable day within the six-week period not affected by substance use. This day typically matched the day of the week of the SUED, was not during the first two weeks of monitoring and excluded atypical exercise or meals.
This design enabled each participant to act as their own control. The SG data analysed were from the 24-h period from the commencement of substance use compared with 24-h period from the matched control day.

| Statistical analysis
We analysed each 24-h period using the glucometrics recommended by the Juvenile Diabetes Research Foundation Artificial Pancreas Project Consortium 23 to compare SUEDs and non-SUEDs. Alcohol use was divided into two groups according to intake of <or ≥40 g. As stimulants were always consumed with ≥40 g of alcohol, this combination formed a separate group. The primary outcomes were the difference in mean SG level, standard deviation and minutes/24 h period out of range (SG < 3.9 mmol/L or >10.0 mmol/L). Secondary measures included mean differences in SG < 3.9 mmol/L, SG > 10 mmol/L, and mean number of FGM scans ( Figure 1). For statistical analysis, transformation of two outcomes was required: a log transformation with a 10-point location shift was applied for minutes/24 h of SG < 3.9 mmol/L, and an inverse square root to the number of FGM reader scans.
For each outcome, two types of general linear mixed models were fitted. A main effects model considered SUEDs as the primary explanatory variable of interest, but included age, gender and HbA1c as covariates. In the interaction model, the primary effect of interest was the interaction of SUEDs and HbA1c but included age and gender. The test statistics reported are for the primary effects of interest in each case. All models included participant and day pair as random effects. For the main effects model, the difference of means is provided; this is the difference in the mean outcome comparing non-SUEDs and SUEDs. For the interaction model, the difference of "slopes" (regression coefficients for HbA1c) according to non-SUEDs or SUEDs is provided.
Both the main effects and interaction models were applied to non-SUEDs and all SUEDs. These models were then applied to analyse alcohol consumption effects alone comparing non-SUEDs, days after <40 g alcohol were consumed, and days ≥40 g alcohol were  Table 2. Models for the third grouping are included in Appendix 1.

| RE SULTS
A cohort of 20 participants was recruited with two withdrawing prior to completion (one lost to follow up and one hospitalized with diabetic ketoacidosis during the 6-week period). In both participant withdrawals, sufficient data had been collected to analyse at least one substance using event. The demographic and clinical characteristics of participants are described in Table 1.
There were no mean differences detected in the primary outcome measures between SUEDS and non-SUEDs for any substance group. Raw numerical data comparing glucometrics between non-SUEDs (n = 61) and SUED days for alcohol <40 g (n = 17), alcohol ≥40 g (n = 31) and stimulants and ≥40 g alcohol (n = 8) are shown in Figure 1 and the main effects and interaction models are in Tables 2   and 3.

Statistically significant differences were observed between
SUEDs and non-SUEDS in the relationship between HbA1c and mean SG, time out of range (min/24 h) and time with SG > 10 mmol/L (min/24 h). A significant difference was also detected in this relationship on days when ≥40 g of alcohol was used (compared with non-SUEDs) for mean SG values but not for other outcome measures.
The above relationships are represented graphically in Figure 2 with the Pearson coefficient (r) representing the strength of the relationship between HbA1c and the outcome. As shown in the Tables 2 and   3, and Figure 2, the relationship between HbA1c and glucometrics was weaker on SUEDs than non-SUEDs.

| DISCUSS ION
This study is the first to record glucose outcomes following alcohol and drug use in a naturalistic context over a 6-week period using FGM. We did not find a direct relationship between glucose parameters and substance use. However, our data showed a weakening of the relationship between HbA1c and a range of glucose parameters on SUEDs versus non-SUEDs. In most studies from the literature, a linear relationship exists between HbA1c and glycaemic control parameters such as mean SG and time out of range. 24,25 We found that HbA1c was less closely associated with mean SG, time out of range (SG < 3.9 and >10 mmol/L) and time with SG > 10 mmol/L on days when substances were used. Also, a dose-dependent effect was observed for those days when alcohol alone was consumed. Regardless of baseline HbA1c, participants had less predictable glucose levels on the days when they consumed ≥40 g of alcohol, than when they consumed <40 g of alcohol or when no alcohol was consumed.
We did not detect a difference in hypoglycaemia rates between Our study findings therefore concur with the literature and suggest that a threshold level of alcohol consumption is relevant. 9,15 Our study did not establish an impact on SG when stimulants were used concurrently with alcohol consumption. However, the number of events available for analysis in our sample was small (n = 8), limiting any definite conclusions. Interpretation of effects of cannabis on SG was not possible because of small numbers of cannabis users (n = 4). While they are no comparable studies including illicit drug, these results would be expected to vary internationally as Australia has a larger prevalence of methamphetamine users (and correspondingly lower numbers of cocaine users) than Europe, UK or USA. Trends to legalize cannabis in North and South America are also likely to impact. This is the largest, and to date the longest, naturalistic study of the effect of alcohol and illicit drugs on glucose levels in young adults with T1D. We identified that, regardless of baseline glycaemic control, glucose levels are less predictable on days when substances are

F I G U R E 1 Boxplots of glucose outcome measures by type of substance use. # Other includes 2 episodes cannabis use and 3 episodes cannabis and stimulant use
used. This suggests that young adults with T1D need to be provided with health promotion messages that highlight this information and be encouraged to intensify glucose monitoring when drinking alcohol and/or using illicit drugs irrespective of baseline glycaemic control.
Larger real-world studies, particularly on currently illicit drugs, are required as the patterns of substance use change both in Australia

TA B L E 2 Main effects and Interaction models for Non-SUEDs -all SUEDs
and internationally. Research is also required to clarify the most effective health messages for this cohort, as well as the role of closed loop insulin delivery systems and FGM to facilitate harm reduction.

| Limitations
The naturalistic setting provides this study with both strengths and weaknesses. It more closely reflects the social context in which substances are consumed and, thus, the outcomes are likely to be more clinically relevant for young adults with diabetes. However, outside of a laboratory environment, it is difficult to control for wider variables likely to affect glucose parameters and this study may, thus, have underestimated the direct metabolic effects of substance use. As an example, as stimulants are rarely used without alcohol in real-world settings, differences related to stimulant use only may not have been detected. Furthermore, as participants were selecting their own beverages, we could not analyse the effects of different alcohol types (eg. white wine vs red wine vs beer) as these were frequently mixed. The metabolic effect of different alcohol types on BGLs may be important.

TA B L E 3
Main effects and Interaction models for non-SUEDs -<40 g alcohol -≥40 g alcohol RJM and YB were involved in trial conceptualization, methodology, supervision and reviewing and editing the first and final draft of the manuscript.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.

A PPEN D I X 1
This appendix provides the analyses for the third operationalisation of SUDS: no substance use, <40 g of alcohol (only), 40 g of alcohol or more (only), 40 g of alcohol or more and other stimulant use. This analysis focussed on the effects of stimulants, and so excluded days on which participants took: cannabis only, cannabis and stimulants without alcohol, or cannabis and alcohol.