Trajectories of Alcohol Consumption Following Liver Transplantation

Abstract

Any use of alcohol in the years following liver transplantation (LTX) approaches 50% of patients transplanted for alcoholic liver disease (ALD). We collected detailed prospective data on alcohol consumption following LTX for ALD to investigate ongoing patterns of use. Using trajectory modeling we identified four distinct alcohol use trajectories. One group had minimal use over time. Two other groups developed early onset moderate to heavy consumption and one group developed late onset moderate use. These trajectories demonstrate that alcohol use varies based on timing of onset, quantity, and duration. Using discriminant function analysis, we examine characteristics of recipient’s pre-LTX alcohol histories and early post-LTX psychological stressors to identify the profile of those at risk for these specific trajectories. We discuss the relevance of these findings to clinical care and preliminarily to outcomes.

Most reports of alcohol use after liver transplantation (LTX) identify the time to specific alcohol use outcomes (e.g. time to first drink) (1,2) or set thresholds of use by which drinking groups are categorized (e.g. those who drank 140gms of ethanol/week) (3,4,5). We also reported on time to first alcohol use, binge use, and frequency of use in our alcoholic liver disease (ALD) LTX recipients (6). We found those who take a drink can quickly transition to a binge (6 drinks) episode, with binge drinking occurring within the first year for over 40% of those who drank (6). Additionally, pre-LTX alcohol history characteristics (diagnosis of alcohol dependence, short sobriety, alcoholism in a 1° biologic relative, other substance use, and addiction rehabilitation) predicted time to these specific thresholds of use.

There is great clinical and therapeutic utility in identifying the timing of onset of specific thresholds of alcohol use, specifically for planning appropriate timing of interventions. Over the long-term, however, whether drinking is sustained, for how long, and whether specific patterns of use emerge may be most important for predicting outcome. Thus as the next step in our investigations of alcohol use in ALD LTX recipients we collected detailed, prospective post-LTX alcohol consumption data allowing us to identify patterns of use far beyond basic thresholds of use. Using group-based trajectory models, designed to identify clusters of individuals who follow a similar progression in behavior or outcome over time (7), we modeled alcohol use based on the quantity, frequency, and duration of consumption.

Our second step was to examine the contribution of both pre- and early post-LTX factors to these trajectories of drinking. Alcohol use is highly influenced by personal risk factors such as genetic predisposition, as well as environmental and psychological factors (8,9). Thus we hypothesized that transplant-specific events would be crucial to both the initiation and eventual pattern of alcohol use. As alcohol use can begin early post-LTX (10) we collected data on psychological, and medical stressors within three months post-LTX. We hypothesized these stressors would be associated with specific trajectories of alcohol use (i.e., those who experienced more stress early on would resume quickly and drink heavily). Using discriminant function analyses we identified key predictors of those who would advance along specific trajectories of alcohol use.

Methods

Subjects

At the Starzl Transplant Institute (STI) from May 1998 to August 2004, 265 patients underwent LTX for a primary or secondary diagnosis of ALD. We used standard research procedures for approaching and obtaining signed informed consent according to our IRB approved study protocol. During the period of recruitment, 208 (78%) of the 265 ALD LTX recipients participated; 38 (14%) died before enrollment, 4 (2%) were ineligible due to prolonged hospitalization/nursing home placement, and 15 (6%) refused to participate.

The pre-LTX diagnosis of ALD was determined by consensus from interviews and examinations by our transplant surgeons (PF and MdV), hepatologists, and psychiatry team (AD and MGF). Patients with ALD had prior excessive alcohol use, defined as ≥ 20 grams of ethanol/day for women or ≥ 60 grams ethanol/day for men (11). The majority (>80%) had consumed these amounts for ≥10 years. Psychiatric diagnoses of alcohol dependence or abuse were made by the psychiatry team using a structured interview and the Diagnostic and Statistical Manual of Mental Disorders IV (12) criteria.

Pre-transplant alcohol and other psychosocial variables

Demographic and psychosocial information collected during the pre-transplant evaluations and documented on a standardized evaluation form was extracted from the medical record including: pre-transplant length of sobriety, attendance at alcohol rehabilitation, family history of alcoholism (patient-reported 1° biologic relative with an alcohol use disorder), history of other substances and injected drug use, and psychiatric diagnosis and treatment for substance use, depressive or anxiety disorders.

Pre- and post-transplant medical variables

Medical variables including the patient’s LTX MELD score, hepatitis C (HCV) infection, post-LTX biopsy data, and causes of death were collected from the medical record. Additional data were collected on episodes of biopsy proven acute rejection. All tissue pathology data are reviewed by trained evaluators in our STI Transplant Pathology department. Acute rejection required both a pathologist’s descriptive report of acute rejection and a Rejection Activity Index score ≥ 3.

Procedures

Interviews and questionnaires

Three prospective measures of post-transplant alcohol use were obtained during either face-to-face research assessments at a return clinic visit, by telephone with the research staff, or by mail. First, every 3 months for the first post-transplant year and every 6 months thereafter, patients completed the Alcohol-Timeline Follow-back questionnaire (ATLFB) (13) (97% by mail). The ATLFB is a calendar instrument used in alcohol research that captures a daily profile of alcohol use for the intervals between follow-up interviews providing information on the quantity, frequency, patterns, and duration of use. The ATLFB has good psychometric characteristics (high test-retest reliability and validity across clinical and general population samples) (14). Participants were told the ATLFB was strictly confidential and would not become a part of their medical record or be revealed to any LTX team member. We felt that insuring participants’ anonymity would improve the yield in reporting alcohol use. Completion rates were high at all time points (range 75% – 94%, average 81% with >81% average at five years and beyond).

Second, over the same time points, a caregiver who knew the patient best and typically lived with the patient (usually a spouse or family member) completed an alcohol quantity–frequency questionnaire that asked about the patient’s alcohol use. The caregiver questionnaire patterned after the NIAAA Quantity-Frequency measure (15) asks about the number of drinking days and the amounts consumed. This information was transcribed onto a calendar form similar to the ATLFB.

Third, during routine post-transplant clinic appointments, clinical interviews were performed by the transplant psychiatrist (AD) who was blinded to the data obtained by other methods. Responses to questions about alcohol use from the psychiatrist’s interview were compared with information given by the patient to the transplant coordinators and surgeons during the same clinic visit in independent interviews. The highest amount reported by the patient was recorded as quantity/frequency of alcohol use with specific dates and amounts of use on a monthly calendar form (similar to the ATLFB). Patients are seen in the transplant clinic as medically indicated. However, when possible, most patients are seen twice weekly for the first month after LTX discharge, then monthly until 3 months post-LTX, then typically every 3 to 6 months thereafter (see reference 10 for comparison of alcohol reporting methods) (10).

Biochemical markers

As part of routine clinical care, blood alcohol levels are checked at transplant clinic appointments. These data were extracted from patients’ medical records and used in combination with alcohol calendars to identify the onset of use (if discrepant from other methods) or to quantify the amount used.

Measures of psychological health and general well-being

At 3 months post-LTX or later if still hospitalized, we assessed symptoms of depression (Beck Depression Inventory-BDI), anxiety (Zung Anxiety Scale-ZAS), perceived stress (Cohen’s Perceived Stress Scale -PSS) and health-related quality of life (SF36). Several items reflecting transplant-specific concerns (16,17) were also administered: patients were also asked to rate how frequently they experienced worries about their health, possible rejection episodes, need for biopsy, need for a second transplant, and regrets about undergoing LTX. They were also asked whether they felt confident they would get another transplant if needed.

Post-LTX Clinical Care

At follow-up clinic appointments patients are counseled to maintain complete abstinence from alcohol. If alcohol use is reported, the patient would receive further counseling from the transplant psychiatrist with referral to professional alcohol counseling if indicated. Thus the study was not a naturalistic study in which alcohol use was merely observed without clinical action being taken. Instead we practiced the same standard of care as would most transplant programs where complete sobriety is recommended and clinical intervention is provided when alcohol use is discovered.

Our standard immunosuppressive protocol is tacrolimus with steroids, mycophenolate mofetil, or sirolimus as indicated. We attempt to decrease the tacrolimus levels to as low as possible for monotherapy maintenance.

Statistical Analyses

Data from the three alcohol use calendar methods (patient, caregiver, clinic reports) were collapsed into a single drinking outcome for each individual. When there was a discrepancy between consumption amounts or frequency we chose the highest reported amount to represent alcohol use assuming that a patient/caregiver would not err on the side of over-reporting. The correlation between the alcohol use utilized in the analyses and the various methods was highest for the clinic report (0.96), although patient report (0.73) and caregiver report (0.47) contributed some cases to the analyses. Importantly the patients’ reported use on their self-report calendar and what they revealed during clinic (r = 0.60) interviews indicated moderately good concordance (18). For purposes of comparison between types of alcohol consumed, alcohol consumption was reported in standard alcohol drink units (i.e. one 12 ounce beer, 5–6 ounces of wine, or a one ounce “shot” of hard liquor).

Trajectory modeling is a latent class analysis which identifies homogeneous groups within a population assumed to contain different patterns or trajectories. To examine patterns (trajectories) of the daily amounts of alcohol consumption over time, the SAS procedure PROC TRAJ (19) (refer to http://www.andrew.cmu.edu/user/bjones) was used. This procedure builds trajectories representing likelihood of and level of alcohol use for the different latent groups as a function of time from baseline. Parameters are estimated via Maximum Likelihood estimates. Here, the trajectories of the total number drinks per day (collapsed into 3 week periods of consumption reported from the point of discharge from the LTX hospitalization) were modeled by a censored normal distribution. Since the models can accommodate data from individuals with varying lengths of follow-up, participants who drop out or died over the course of follow-up do not need to be excluded; each participant’s data until the point of attrition were included. Model search involves fitting models and increasing the number of groups in the models until the maximum Bayesian Information Criteria (BIC) (20) is identified, signifying the optimal number of homogenous groups. Polynomial functions (up to the fifth order) in time post-LTX are considered for each group.

We used discriminant function analysis to 1) determine whether pre-LTX alcohol use history variables as well as participants’ medical and psychosocial characteristics early post-LTX could reliably discriminate between trajectory groups and 2) to assess the relative importance of these predictors in accurately classifying individuals into the groups. Due to the small group sizes, the full set of variables could not be used in the discriminant analysis and a subset of ten variables were chosen. First, based on our prior clinical experience and research (6), we chose four pre-LTX variables a priori from the alcohol history to examine: length of sobriety, psychiatric diagnosis of alcohol dependence, 1° biologic relative with alcoholism, and use of other substances (addiction rehabilitation was not used due to high collinearity with other variables). Second, we included variables reflecting medical status: pre-LTX MELD score and early post-LTX physical complaints (pain, low energy/fatigue). (Co-infection with HCV was not used due to high collinearity with other substance use) Third, we considered the contribution of symptoms of affective distress and stress (ZAS, BDI, PSS) and chose the perceived stress scale, a broad measure of overall life stress, due to significant collinearity between these measures. We also were interested in perceptions of the transplant experience and felt questions on whether recipient’s would chose to undergo transplant, knowing what they now know, and whether they had confidence about receiving another transplant if something happened to their current liver best captured the early post-LTX satisfaction (or dissatisfaction) with their experience.

One hundred fifty nine recipients (76%) had complete data on all of the variables and were included in the discriminant function analysis. Participants without all variables joined the study later when the psychological measures were not being collected. These individuals were similar on demographic, alcohol, psychiatric, medical and psychological variables except that they were more likely to be single and have histories of other substance use.

Results

Pre-LTX History and Characteristics (Table 1)

Table 1.

Cohort Demographic, Psychiatric, and Medical Characteristics

Demographics
Gender, % male	85
Age, M (SD)	50 (8)
Race, % white	93
Education, % > high school	51
Occupation, % blue collar	75
Marital status, % married	45
Alcohol and Psychiatric History
Years heavy drinking, M (SD)	20 (9)
Months of sobriety, M (SD)b	39 (47)
Average drinks/week, M(SD)b	107 (109)
Alcohol rehabilitation, % yes	48
Family history of alcoholism, % yes	62
Alcohol dependence diagnosis, % yes	79
Other substance use, % yes	42
Depressive disorder, % yes	19
Medical Variables
Hepatitis C infection, % yes	51
MELD score, M (SD)b	16 (7)
Acute rejection within 3 months, % yes	27
SF 36 Bodily Pain scale	53 (25)
SF 36 Vitality scale	48 (21)
Early Post-LTX Psychiatric Variables
Perceived stress Scale, M (SD)	19 (9)
Beck Depression Inventory, M (SD)	11 (7)
Zung Anxiety Scale, M (SD)	35 (8)

^aχ²(4) for dichotomous variables, F (4,159) for continuous variables.

^blog transformed to reduce skewness prior to statistical test. Untransformed means and SDs are presented to facilitate interpretation.

Cohort Demographics

Demographic characteristics of are cohort are similar with respect to age, gender, marital status, and race to national statistics for ALD LTX recipients (21). Patients are predominantly European American (93%), male (85%), married (45%), with a mean age 52±7 years. Most were high school educated and worked in non-professional jobs (Hollingshead classification).

Pre-LTX Alcohol Histories

Prior to LTX most patients had 20+ years of heavy drinking although there were gender differences (mean 20±9 years for men vs. 14±8 for women, p=0.001). However daily/weekly amounts of alcohol consumed did not differ by gender (average 15 standardized drinks/day). Sixty-two percent had a 1° biologic relative with alcoholism. Most had ≥ year of pre-LTX sobriety, but only 48% had attended addiction rehabilitation. In addition to alcohol, 45% had used other drugs including marijuana, cocaine, and heroin.

Medical Variables

HCV infection was present in 51% of patients (Table 1). The mean MELD score at the time of LTX was 16. Following LTX, 27% of the patients experienced acute rejection within three months and 8% had two or more episodes of rejection.

Post-LTX Characteristics and Outcomes

Psychological health, general well-being, and perceptions of LTX

While the 3-month post-LTX mean scores for the ZAS (35.0), BDI (11.4), and PSS (18.7) are not indicative of clinically significant levels of anxiety, depression, or stress, 20–25% of patients reported mild-severe symptoms on each of these measures. Health-related quality of life (mean SF36 subscale scores) are significantly and expectedly worse compared to the general non-patient populations for general health (60.7), physical functioning (56.2), role physical (28.8), role emotional (69.1), social functioning (68.1), vitality (48.2), mental health (72.1), and bodily pain (52.9). U.S. non-patient population norms for the SF36 (range 0–100) for subscale scores are in the high 70s and above. Nonetheless, 96% did not regret undergoing LTX. Fifty-seven percent of the patients were worried about a possible rejection episode and over 50% had concerns about their health and worried about needing a new liver. However, 76% felt confident that they would get another LTX if they needed one.

Alcohol Use Trajectories

Of the 208 patients, 113 (54%) had no reported alcohol use post-LTX on any measure (group 1). Within the 95 non-abstainers, we identified four distinct trajectories of alcohol consumption. The majority (n=55, group 2) drank low amounts infrequently. However three other distinct alcohol use trajectories emerged: early onset moderate use that diminished over time (n=13, group 3), later onset moderate use that increased over time (n=15, group 4), and an early onset, heavy, increasing pattern of use (n=12, group 5).

The trajectories (figure 1) are displayed as the natural log of the total number of standard drinks for three week intervals to highlight the shapes of the curves with confidence intervals displayed around each trajectory line. The amounts of consumption reflect total consumption during each of the three-week sequential time periods over a period of multiple years post-transplant. Our sample size did not allow examination of use more precisely than overall consumption during these time segments.

Figure 1.

Specific Alcohol Use Trajectories From the Point of Transplant

Table 3 describes the characteristics of each trajectory. Group 2 had minimal fluctuating use throughout the nine-year period. Group 3 had a rapidly accelerating moderate use pattern that peaked at 1.7 years and began decreasing by year 3. Group 4 had minimal use until year 3 then steadily increasing use that peaked during the sixth year. Group 5, the heaviest drinking group, had a rapid onset increasing heavy use that peaked at year 3 and slowly decreased through year 7. Although the average across group 5 was 3.5 drinks/day, there were individuals in this group who consumed, at times, up to a six pack of beer or a fifth of hard liquor/day.

Table 3.

Specific Characteristics of Alcohol Use Trajectories

Group	Onset of Use After LTX Discharge	Pattern Of Use	Average Consumption at Peak Amount*	Timing of Heaviest Use
Group 1	None	Complete abstinence
Group 2	2.8 months	Fluctuating low level of use	½ standard drink/week
Group 3	3.5 months	Early onset rapidly accelerating moderate use	3.5 standard drinks/week	Peaked 1.7 years
Group 4	2.8 months	Steady increase to moderate use after 3 years post-LTX	2 standard drinks a day	Peaked at year 6
Group 5	42 days	Early onset continuously increasing heavy use	3.7 standard drinks a day	Peaked at year 3

^*For purposes of comparison between types of alcohol consumed, alcohol consumption is converted into standard drink units (i.e. one 12 ounce beer, 5–6 ounces of wine, or a one ounce “shot” of hard liquor).

We preliminarily explored outcomes between the groups. We expected that alcohol exposure would have to be significant to impact outcomes, thus we chose the two early onset moderate to heavy use groups (groups 3 and 5) to compare to all others. The heavier use of group 4 did not occur until beyond three years and would not have such a strong effect on early outcomes. We found groups 3 and 5 differed from all others by more frequently having steatohepatitis or rejection on biopsy and being more likely to experience graft failure. All who died of recurrent alcoholic liver disease were in groups 3 and 5. There was no difference in HCV co-infection between groups nor were those in groups 3 and 5 more likely to die of recurrent HCV (see Table 4).

Table 4.

Comparison Of Outcomes Between Early Onset Drinkers Vs. Others

Outcomes	Groups 3 and 5	All Others	Exact p*
≥20% of biopsies with steatohepatitisa	23%, n=5	9 %, n=15	p=0.05
≥40% of biopsies with acute rejectionb	41%, n=9	18%, n=32	p=0.02
Graft Failure	73%, n=8	37%, n=25	p=0.04
Causes of Death	N=11c	N=68
Alcoholic liver disease	46%, n=5	0%	p=0.00
Recurrent HCV	36%, n=4	28%, n=19	p=0.72
Other causes of liver failured	9%, n=1	10%, n=7	p=1.00
Malignancye	0%	21%, n=14	p=0.19
Cardiac eventsf	18%, n=2	10%, n=7	p=0.60
Infection or sepsis	9%, n=1	12%, n=8	p=1.00
Unknown	0%	10%, n=7	p=0.58
Accident/trauma	0%	4%, n=3	p=1.00
Other	0%	4%, n=3	p=1.00

^*Exact p values are reported because the numbers of cases are small (2-sided Fisher’s exact p)

^aTen participants did not get biopsies

^bRejection Activity Index score ≥ 3.

^cTotal number adds to more than 11 as two persons died of a combination of recurrent HCV and alcohol related liver disease

^dIncludes PNF, HAT, PVT, infarction

^eIncludes hepatocellular carcinoma, lung, and skin

^fIncludes myocardial infarction, heart failure, arrhythmia

Predictors of alcohol use

We used multivariate discriminant function analysis to evaluate whether recipients in the five trajectories could be distinguished from each other across an array of interrelated factors representing pre-LTX characteristics and post-LTX stressors. The discriminant function analysis compared all five trajectories on a final subset of these ten characteristics (see Table 2.). The trajectory groups varied along two dimensions (“functions”) in the discriminant function analysis. Each dimension accounted for a significant portion of the predictors’ discriminating power. A total of 70% of the variability was accounted for by these two functions.

Table 2.

Discriminant Function Analyses using Alcohol History, Psychological Characteristics, and Early Perceptions of Transplantation

	Alcohol Use Trajectories					Discriminant Function Loadings
	Abstainers (Group 1)	Low level drinkers (Group 2)	Early onset moderate drinkers (Group 3)	Late onset moderate drinkers (Group 4)	Early onset heavy drinkers (Group 5)	Dimension 1	Dimension 2
Alcohol and Psychiatric history
Months of sobriety, Ma	47(54)	36(42)	27(25)	27(28)	15(8)	−0.61	−0.11
Family history of alcoholism, % yes	56	66	62	60	92	0.38	0.02
Alcohol dependence diagnosis, % yes	75	85	85	73	83	0.35	0.16
Other substance use, % yes	39	42	77	47	33	0.10	0.25
Early Post-LTX quality of life and psychological symptoms
Bodily pain, M(SD) (high=less)	52 (25)	55 (26)	55 (22)	60 (29)	41 (24)	0.13	−0.26
Vitality M (SD) (high=more)	46 (21)	51 (19)	38 (23)	57 (26)	47 (18)	0.21	−0.47
Health compared to 1 yr ago, % worse	4	4	17	14	18	−0.05	0.72
Confident about re-LTX, % yes	78	75	60	92	64	−0.07	−0.56
Would not get LTX if knew what know now, % yes	6	2	9	0	0	0.24	0.18
Perceived Stress Scale M (SD)	19 (9)	17 (7)	22 (10)	18 (10)	19 (7)	−0.10	0.36

^alog transformed to reduce skewness prior to statistical test. Untransformed means and SDs are presented to facilitate interpretation. Before removal of either function, χ²(40, n=159) =75.3, p=0.001. After removal of the first function, χ²(27, n=159) =40.4, p=0.04. After removal of the second function, only nonsignificant discriminating power remained (χ²(16, n=159) =24.3, p=0.08).

The first function reliably discriminated between complete abstainers and those who drank. The second function maximally discriminated those who had early onset of moderate to heavy alcohol use from those who drank in lesser amounts or accelerated later post-LTX. Group 1 (abstainers) was in a neutral location on this dimension (see Figure 2 for the configuration of group functions and centroids).

Figure 2.

The discriminant function loadings are shown in the last two columns of Table 2. These are interpretable in the same manner as factor loadings in factor analysis and indicate that the most important risk factors, those with loadings at or exceeding 0.25 for the first function (alcohol users vs. abstainers), were shorter length of sobriety (loading=−0.61), positive family history of alcoholism (0.38), and alcohol dependence diagnosis (0.35).

For the second function those with early onset moderate to heavy alcohol use (groups 3 and 5) were distinguished by reporting poorer post-LTX health compared to one year prior (0.72), having more bodily pain (−0.26), feeling less energetic/more fatigued (−0.47), having more perceived stress (0.36), having a pre-LTX history of other substance use (0.25), and not feeling confident that they would get another liver if they needed it (−0.56).

Group Classification

The utility of these discriminant functions can be examined via their ability to correctly classify each participant into their actual trajectory group. One hundred sixteen (60%) were classified correctly into their respective trajectory group. Group-specific classification accuracy was 85%, 48%, 30%, 15%, and 18% for groups 1–5 respectively. This was substantially better than by chance: accuracy by chance would be 49%, 29%, 6%, 8%, and 7%, respectively. The 60% of the total group correctly assigned was also substantially better than 35% by chance alone.

Discussion

Beyond the mere identification of alcohol use, our trajectories highlight important patterns that emerge along the long-term post-LTX course. While most patients abstain or drink minimally, three distinct patterns of consistent moderate to heavy consumption emerged. Two were patterns of early onset use, one moderate and one accelerating to heavy use. These trajectories demonstrate that for some patients, resumption occurs early following transplantation and recipients can quickly lose control over their drinking. For others the resumption of moderate to heavy alcohol use can begin years post-LTX. Thus clinical monitoring should extend well beyond the early years post-LTX.

We confirmed, as in our prior reports (6), that pre-LTX variables of the alcohol history (alcohol dependence, short length of sobriety, family history of alcoholism and the use of other substances) identifies drinkers compared to abstainers. Also, as in our prior analyses, length of sobriety is the most powerful predictor of return to alcohol use with the highest discriminant factor loading.

The discriminant function analyses identified additional characteristics of those who would return to a moderate or heavy pattern of alcohol use. While it might be expected that early post-LTX health concerns would prevent or forestall a return to alcohol use, we hypothesized that stresses immediately following LTX would increase the risk of alcohol use. We found that those more likely to drink in the early problematic patterns (groups 3 and 5) were experiencing more problems overall, were more stressed, reported worse health, had more pain and less energy, and felt less confident they would get another liver if they needed it. Perhaps it was a sense of the ineffectualness of their situation: the immediate result of the transplant did not more quickly transform them to better health, they felt their health was worse and that they felt they would not get another liver. An alternative explanation could be that alcohol use was how they handled stress, and they gravitated to drinking in response to the difficulties of the early post-LTX phase. Early identification and treatment of stress especially as it relates to early post-LTX recovery, attention to complaints of pain and fatigue, as well as resumption of addiction counseling may aid in the stabilization of these patients.

Those in groups 2 and 4 who drank minimally and consistently throughout the early post-LTX years, had characteristics opposite of groups 3 and 5. They were experiencing better health than a year ago, were less stressed, less likely to be in pain, felt more vital, did not regret their decision to undergo LTX, and felt more confident they would get another liver if needed. Thus, those patients who could be considered as doing very well in the early stage post-LTX may be more complacent about their overall health and the health of their liver and feel that occasional drinking early on is not a problem.

A possible limitation of the study is the small size of our trajectory groups. However, to consider alcohol use beyond just dichotomous outcomes we needed to consider smaller groups from which these distinct patterns of alcohol consumption emerged. As a single center study the sample size was not large enough to perform cross validation. Future work in larger LTX cohorts is needed both to ensure that the trajectories are reliable and to allow for a greater number of potential predictors to be examined. In the present analyses, we limited the number of predictors examined to ensure that the effects could be adequately examined (22). Additionally participation in the study may have changed behavior. This was not a naturalist study as noted above.

Although we have preliminarily examined early outcomes between groups, in future work we plan to examine these trajectories in relation to subsequent late onset medical events. As we have created trajectories using the ten-year period of follow-up, we will require further longitudinal outcome data (beyond these ten years) to determine whether the trajectories precede and predict medical events.

Thus we have found striking patterns in the evolution of alcohol consumption that emerge after LTX for ALD. These patterns demonstrate that alcohol use can be very varied beginning early or later, and can stay at minimal levels or accelerate into heavy sustained patterns of use. We also importantly identify the predictors of those who will abstain vs. drink and also those who will advance along specific trajectories. From these findings we can now target our preventative and interventional treatment resources to these specific individuals at risk at specific time points of risk.