Longitudinal trajectories of substance use disorder treatment use: A latent class growth analysis using a national cohort in Chile

Abstract

Background and aims:

Longitudinal studies have revealed that substance use treatment use is often recurrent among patients; the longitudinal patterns and characteristics of those treatment trajectories have received less attention, particularly in the global south. This study aimed to disentangle heterogeneity in treatment use among adult patients in Chile by identifying distinct treatment trajectory groups and factors associated with them.

Design:

National-level registry-based retrospective cohort.

Setting and participants:

Adults admitted to publicly funded substance use disorder treatment programs in Chile from November 2009 to November 2010 and followed for 9 years (n = 6266).

Measurements:

Monthly treatment use; type of treatment; ownership of the treatment center; discharge status; primary substance used; sociodemographic.

Findings:

A seven-class treatment trajectory solution was chosen using latent class growth analysis. We identified three trajectory groups that did not recur and had different treatment lengths: Early discontinuation (32%), Less than a year in treatment (19.7%) and Year-long episode, without recurrence (12.3%). We also identified a mixed trajectory group that had a long first treatment or two treatment episodes with a brief time between treatments: Long first treatment, or immediate recurrence (6.3%), and three recurrent treatment trajectory groups: Recurrent and decreasing (14.2%), Early discontinuation with recurrence (9.9%) and Recurrent after long between treatments period (5.7%). Inpatient or outpatient high intensity (vs. outpatient low intensity) at first entry increased the odds of being in the longer one-episode groups compared with the Early discontinuation group. Women had increased odds of belonging to all the recurrent groups. Using cocaine paste (vs. alcohol) as a primary substance decreased the odds of belonging to long one-episode groups.

Conclusions:

In Chile, people in publicly funded treatment for substance use disorder show seven distinct care trajectories: three groups with different treatment lengths and no recurring episodes, a mixed group with a long first treatment or two treatment episodes with a short between-treatment-episodes period and three recurrent treatment groups.

INTRODUCTION

Substance use disorders (SUDs) are often characterized as chronic and reoccurring disorders [1–6]. Individuals with SUDs exhibit distinct stages such as heavy use, treatment utilization, relapse, and recovery or remission. Some patients recover without formal intervention [7, 8], whereas others engage in peer-based supports (i.e. alcoholic anonymous) and/or formal treatment to achieve or maintain remission [8–11]. Hence, individual paths and trajectories are highly heterogeneous.

When patients do seek treatment for SUDs, their care trajectories often vary by time in treatment, type of treatment (e.g. inpatient or outpatient), number of episodes and outcomes. Multiple readmissions are fairly common [6, 12–19], with 20% to 80% of participants reporting previous treatments [6, 12, 20–22]. Understanding treatment trajectories can inform policy by distinguishing groups at risk of discontinuing and those who repeatedly use treatment [23], as well as their influence on current engagement and outcomes. Consequently, governmental agencies and policymakers can improve retention and re-engagement strategies, which are key for improving long-term outcomes among patients who seek treatment, including a reduction in substance use, criminal legal system involvement, fatal and non-fatal overdoses and improvements in social functioning and employment [24–28].

Despite this need, analyzing longitudinal treatment utilization patterns over time is extremely complex [22]. Previous research analyzing treatment utilization usually dichotomizes patients as first-timers or people who had previous treatments [20, 29–31], which oversimplifies the heterogeneity among those with prior treatment. Longitudinal studies are scarce [32] and rely on self-reports [33–35] or are collected retrospectively over broad periods [23]. National registries have been analyzed cross-sectionally [20]. Moreover, most research regarding treatment utilization takes place in high-income countries, where drug supply, patterns of misuse and healthcare delivery differ from the ones in the global South and other low and middle-income regions. In Latin America, both the prevalence of opioid and injection drug use are low [36, 37]. However, cocaine is mostly produced in the Andes mountain range, making tobacco, alcohol, cocaine and cocaine derivates (including cocaine paste, a smokable intermediary product in the extraction of cocaine from coca leaves) [38] the main public health concerns [36].

Previous studies found inconsistent associations between age [19, 20, 31, 39, 40], sex [1, 6, 19, 20, 31] and treatment utilization. People with higher education tend to report more previous treatments and longer stays [1, 19, 31], and unemployment has been linked to recurrence [19, 31]. Being married has been associated with shorter stays in treatment [1], but increased recurrence [31]. Patients with more severe substance use problems and longer substance use histories are more likely to seek treatment multiple times [6, 12, 20, 29–31, 39, 41], and people with multiple treatment episodes exhibit worse mental health [12, 20, 39] and higher involvement in the criminal legal system [6, 20, 41]. Moreover, factors such as type of treatment and geographic region are usually overlooked, although recent research has shown that these factors have a greater impact than individual characteristics in treatment discontinuation [42].

To address these gaps, we aim to compare patients presenting one versus two or more treatment episodes. Later, we identify groups with distinctive patterns of treatment utilization using latent class growth analysis (LCGA) in a rich national-level registry-based retrospective cohort. Since 2010, Chile has had one of Latin America’s most advanced treatment monitoring systems called the Treatment Information System (SISTRAT), managed by the Service for Prevention and Rehabilitation of Drug and Alcohol Consumption (SENDA) [43, 44]. This institution sets treatment standards, finances most treatment for publicly insured people and oversees implementation [45]. As such, these data allow for a unique opportunity to identify groups of patients showing distinctive treatment utilization trajectories and explore geographical, treatment and individual-level factors associated with them.

METHODS

Study design and participants

We used a national-level registry-based retrospective cohort design of adult population admitted to Chile’s SUDs treatment programs from November 2009 to November 2010 (n = 6266) and followed them for 9 years (108 months). The selected data encompasses all admissions and discharges from SISTRAT, which includes publicly funded months of treatment in public and private institutions. In 2010, ~83% of the population had public health insurance, and SENDA is the main institution financing those treatment plans (~80%). There were 9963 treatment episodes in the cohort; however, 258 were combined and treated as a single episode (2.6% of the total). Specifically, when a patient was referred within 45 days to another treatment within the same network because of reasons suggested by the center’s team (such as a change of address or treatment plan) consecutive treatment episodes were counted as one. Therefore, our analysis encompasses a total of 9705 treatment episodes. The Ethics Committee of Universidad Mayor, Chile, reviewed and approved this study (No. 260/2019). The analysis for this study was not preregistered, therefore, the results should be considered exploratory.

Measurements

Monthly treatment utilization

We used each treatment episode’s admission and discharge dates to construct episodes of time in and out of treatment on a monthly scale. Two hundred twenty-six of the 9705 episodes lasted less than half a month (2.3%), nonetheless, we defined that each episode accounted for at least 1 month to address every contact with the system. Between-treatment episode durations that contributed less than half a month were rounded to zero. Therefore, a given participant might present two treatment episodes and be coded with continuous use of services if the time between those episodes was less than half a month.

Treatment characteristics

For each treatment episode, we identified the type, ownership of treatment center and discharge status. Type of treatment was defined as (i) outpatient low intensity; (ii) outpatient high intensity; or (iii) inpatient. SENDA’s technical orientations recommend low intensity programs to implement a 6-month treatment, with three times a week intervention usually during the afternoons for 3 to 5 hours. Outpatient high intensity modality treatments should last 8 months, with five times attendance during the week, usually in the afternoon for at least 5 hours. Inpatient modalities are encouraged to last 1 year [46]. Ownership of treatment centers could be (i) public or (ii) private. The discharge status was defined as (i) early discontinuation (≤90th day in treatment); (ii) late discontinuation (>90th day in treatment); (iii) administrative discharge (because of serious misconduct); (iv) completed treatment; (vi) referral; or (vii) currently in treatment (as of November 2019). Geographic macrozone of the center included (i) north; (ii) center; or (iii) south.

Patients’ baseline characteristics

For each patient, we included (i) urbanicity of the commune of residence (rural, mixed, urban); (ii) sex (women, men); (iii) age group (18–29, 30–39, 40–49, 50 or more); (vi) education (more than high school, completed high school or less, completed primary school of less); (v) employment (unemployed, employed, inactive); (vi) marital status (married/shared living arrangements, separated/divorce, single, widower); (vii) cohabitation (family, others, alone); (viii) number of children (none, one, two, three or more); and (ix) primary substance (alcohol, cannabis, cocaine [powder], cocaine paste, others).

Statistical analysis

Bivariate analysis

We conducted bivariate analyses to compare the characteristics of patients with one versus two or more treatment episodes. We used t tests for continuous variables and χ² tests for categorical variables.

LCGA

For the construction of the trajectories of monthly treatment utilization, we used LCGA using the lcmm R package (for the model’s diagram, see Figure S1 [47]. As part of the finite mixture modeling framework, LCGA assumes that a latent variable, composed of several classes, underlies the heterogeneity in how a given outcome evolves over time [48–50]. Therefore, growth models attempt to estimate between-person differences in within-person change over time. LCGA is a special type of growth mixture model, where the variance and covariance estimates for the growth factors within each class are fixed to zero [48]. By this assumption, all individual growth trajectories within a class are homogeneous. We used a three-step approach: (i) we fitted the latent trajectory model; (ii) assigned the individuals to their most likely class; and (iii) explored correlates [51]. We estimated models ranging from one to eight classes with monthly repeated measures of treatment utilization (yes/no) with intercept, slopes and linear/quadratic parameters as in previous studies [23]. The model fitting was assessed by goodness-of-fit indexes, which included the Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), sample-size adjusted BIC (SABIC), entropy and the integrated completed likelihood criterion (ICL). Lower values of AIC, BIC, SABIC and ICL and higher values of log-likelihood and entropy were expected for a well-fitted model. We also conducted Lo–Mendell–Rubin likelihood ratio test (LMR-LRT) to compare the fit of the models. To inform the model selection, and because LCGA can identify multiple class categorizations [52, 53], we compared our solutions of 7 and 6 classes and assessed if the last included class in the seven-class solution differed in terms of key features, such as length in treatment, recurrence and discharge status (Tables S4–S5). For further information regarding the LCGA, please review Tables S1–S7 and Figures S1–S7. Afterward, we described key features of the groups such as length in treatment, recurrence and discharge status and assessed differences between the groups resulting from the LCGA using bivariate analysis.

Multinomial logistic regression model

We conducted a multivariate multinomial logistic regression to estimate associations between covariates and the groups identified in the LCGA. We used the patients’ baseline characteristics and the type, ownership and macrozone of the first treatment as potential correlates (see Table S8). To account for missing data (0%–1% for five covariates), we used multiple imputations by chained equations (m = 20) using the nnet and the mice packages in R. Unimputed results are shown as imputed results did not show noteworthy changes (see Table S9). The code is available at https://github.com/idborquez/treatment_trajectories.

Sensitivity analyses

To check the stability of our solution with less information (time points) we re-estimated the models of 6, 7 and 8 classes with half the time points (54, every 2 months). Likewise, we re-estimated the model predicting the seven-class solution to assess if the direction and size of the effects identified in our primary analysis changed (Tables S10–S12).

RESULTS

Table 1 displays the characteristics of the patients comparing those with one versus two or more treatment episodes. Among the entire cohort, 73% were men, with an average age of 33.8 years old, and 44.7% were employed at baseline. A total of 65.8% experienced only one treatment episode, whereas 4.9% had four or more. On average, patients in the cohort underwent 1.5 treatment episodes, with 2.6 among people with more than one episode. Those with multiple episodes were more likely women (31.8% vs 24.5%, P < 0.001), under 29 years old (42.6% vs 37.6%, P < 0.001) and unemployed (45.9% vs 40.1%, P < 0.001).

TABLE 1.

Participant characteristics according to the number of treatment episodes (one vs more than one).

	Mean (SD) or n (%)
	No. of treatment episodes
	One (n = 4120)	More than one (n = 2146)	Total (n = 6266)	P-value
Urbanicity of the commune of residence				0.195
N-Miss	0	1	1
Rural	169 (4.1%)	77 (3.6%)	246 (3.9%)
Mixed	232 (5.6%)	102 (4.8%)	334 (5.3%)
Urban	3719 (90.3%)	1966 (91.7%)	5685 (90.7%)
Sex				<0.001
Men	3110 (75.5%)	1463 (68.2%)	4573 (73.0%)
Women	1010 (24.5%)	683 (31.8%)	1693 (27.0%)
Age	34.4 (10.1)	32.7 (8.6)	33.8 (9.7)	<0.001
Age group				<0.001
N-Miss	1	0	1
18 to 29	1547 (37.6%)	915 (42.6%)	2462 (39.3%)
30 to 39	1417 (34.4%)	781 (36.4%)	2198 (35.1%)
40 to 49	796 (19.3%)	366 (17.1%)	1162 (18.5%)
50 or more	359 (8.7%)	84 (3.9%)	443 (7.1%)
Education				<0.001
N-Miss	46	17	63
More than high school	1152 (28.3%)	610 (28.7%)	1762 (28.4%)
Completed high school or less	2229 (54.7%)	1241 (58.3%)	3470 (55.9%)
Completed primary school or less	693 (17.0%)	278 (13.1%)	971 (15.7%)
Employment				<0.001
Unemployed	1653 (40.1%)	985 (45.9%)	2638 (42.1%)
Employed	1946 (47.2%)	854 (39.8%)	2800 (44.7%)
Inactive	521 (12.6%)	307 (14.3%)	828 (13.2%)
Cohabitation				0.026
Family	3401 (82.5%)	1827 (85.1%)	5228 (83.4%)
Others	371 (9.0%)	157 (7.3%)	528 (8.4%)
Alone	348 (8.4%)	162 (7.5%)	510 (8.1%)
Marital status				0.101
N-Miss	7	4	11
Married/shared living arrangements	1482 (36.0%)	715 (33.4%)	2197 (35.1%)
Separated/divorce	504 (12.3%)	249 (11.6%)	753 (12.0%)
Single	2090 (50.8%)	1159 (54.1%)	3249 (51.9%)
Widower	37 (0.9%)	19 (0.9%)	56 (0.9%)
Number of children				0.070
N-Miss	26	12	38
None	1041 (25.4%)	521 (24.4%)	1562 (25.1%)
One	1137 (27.8%)	640 (30.0%)	1777 (28.5%)
Two	1023 (25.0%)	556 (26.1%)	1579 (25.4%)
Three or more	893 (21.8%)	417 (19.5%)	1310 (21.0%)
Primary substance				<0.001
Alcohol	1005 (24.4%)	345 (16.1%)	1350 (21.5%)
Cocaine	691 (16.8%)	361 (16.8%)	1052 (16.8%)
Cocaine paste	1888 (45.8%)	1232 (57.4%)	3120 (49.8%)
Cannabis	411 (10.0%)	162 (7.5%)	573 (9.1%)
Other	125 (3.0%)	46 (2.1%)	171 (2.7%)
Number of treatment episodes	1	2.6 (1.0)	1.5 (0.9)	<0.001
Total months in the treatment system	7.4 (5.8)	60.2 (32.9)	25.5 (31.9)	<0.001
First treatment episode duration (months)	7.4 (5.8)	7.2 (6.2)	7.4 (6.0)	0.211
Geographical macrozone of the treatment center				0.232
Center	3294 (80.0%)	1701 (79.3%)	4995 (79.7%)
North	576 (14.0%)	329 (15.3%)	905 (14.4%)
South	250 (6.1%)	116 (5.4%)	366 (5.8%)
Type of treatment at first entry
Outpatient low intensity	1860 (45.1%)	763 (35.6%)	2623 (41.9%)	<0.001
Outpatient high intensity	1474 (35.8%)	818 (38.1%)	2292 (36.6%)
Inpatient	786 (19.1%)	565 (26.3%)	1351 (21.6%)
Ownership of the treatment center at first entry				<0.001
Public	2852 (69.2%)	1332 (62.1%)	4184 (66.8%)
Private	1268 (30.8%)	814 (37.9%)	2082 (33.2%)
First treatment discharge status				<0.001
N-Miss	0	1	1
Early discontinuation	668 (16.2%)	380 (17.7%)	1048 (16.7%)
Late discontinuation	1712 (41.6%)	841 (39.2%)	2553 (40.8%)
Administrative discharge	446 (10.8%)	279 (13.0%)	725 (11.6%)
Completed treatment	890 (21.6%)	402 (18.7%)	1292 (20.6%)
Referral	404 (9.8%)	243 (11.3%)	647 (10.3%)
Last treatment discharge status				–a
Early discontinuation	668 (16.2%)	344 (16.0%)	1012 (16.2%)
Late discontinuation	1712 (41.6%)	739 (34.4%)	2451 (39.1%)
Administrative discharge	446 (10.8%)	192 (8.9%)	638 (10.2%)
Completed treatment	890 (21.6%)	489 (22.8%)	1379 (22.0%)
Referral	404 (9.8%)	264 (12.3%)	668 (10.7%)
Currently in treatment	0 (0.0%)	118 (5.5%)	118 (1.9%)

Note: P-values were obtained using t test for continuous variables and χ² test for categorical variables. Tests were conducted excluding missing data. 0%-1% of missingness across covariates.

^aP-value not shown for lack of observed counts.

Trajectories of treatment utilization

All fit statistics favored the seven-class solution with quadratic slopes and it presented high entropy (>0.8) (Table 2). The LMR-LRT remained significant when comparing the 6 and 7 class solutions (P < 0.001). The seventh class that emerged corresponded to a one-episode trajectory group (Figures S6–S7), with distinctive characteristics (e.g. in terms of length of treatment and discharge status) from the other one-episode groups (Tables S4–S5). Therefore, based on the overall fit and the interpretability of the trajectories, we selected a seven-class model with quadratic terms. The average posterior probabilities for all groups were above 0.9, and the odds of correct classification were above 5, indicating a good overall fit (Table S6).

TABLE 2.

Fit statistics for 1 to 8 latent classes linear/quadratic solutions (n = 6266).

	Class no.	Convergence	AIC	BIC	Changes in BIC (%)	SABIC	ICL	Entropy	LMR-LRT
Linear	1	Yes	305 727	305 747	-	305 738	305 747	1.00	<0.001
	2	Yes	216 455	216 495	29.2	216 476	204 029	0.97	<0.001
	3	Yes	205 527	205 587	5.0	205 559	193 229	0.96	<0.001
	4	Yes	198 043	198 124	3.6	198 086	186 082	0.93	<0.001
	5	Yes	195 233	195 334	1.4	195 286	183 461	0.92	<0.001
	6	Yes	192 325	192 446	1.5	192 389	180 666	0.91	<0.001
	7	Yes	190 798	190 940	0.8	190 873	179 205	0.92	<0.001
	8	Yes	189 881	190 043	0.5	189 967	178 419	0.91	<0.001
Quadratic	1	Yes	263 906	263 933	-	263 920	263 933	1.00	<0.001
	2	Yes	209 837	209 891	20.5	209 866	197 475	0.96	<0.001
	3	Yes	200 737	200 818	4.3	200 779	188 505	0.95	<0.001
	4	Yes	194 426	194 534	3.1	194 483	182 517	0.92	<0.001
	5	Yes	188 960	189 094	2.8	189 031	177 159	0.92	<0.001
	6	Yes	186 393	186 555	1.3	186 479	174 674	0.92	<0.001
	7	Yes	184 484	184 673	1.0	184 584	172 848	0.92	<0.001
	8	-	-	-	-	-	-	-	-

Note: For more information see Table S1.

Abbreviations: AIC, Akaike information criterion; BIC, Bayesian information criterion; ICL, integrated completed likelihood criterion; LMR-LRT, Lo-Mendell-Rubin likelihood ratio test; SABIC, sample-size adjusted BIC.

Figure 1 depicts each trajectory group and the probability of treatment utilization throughout time. The seven classes were characterized as follows: (i). ‘Early discontinuation’ (32%): this group spent an average of 3 months in treatment (SD = 1.3, median = 3). Almost all (98%) had only one episode and 3.4% were discharged with treatment completion. (ii) ‘Less than a year in treatment’ (19.7%): this group spent an average of 8.9 months in treatment (SD = 1.5, median = 8), with one episode (96.4%). Their discharge status was coded mostly as discontinuation or administrative discharge (65.4%), although a quarter (26.7%) had completed treatment. (iii) ‘Year-long episode, without recurrence’ (12.3%): this group spent an average of 13.5 months in treatment (SD = 2.6, median = 13). Again, 97% had only one episode, and almost half (49.2%) had completed treatment as discharge status. (iv) ‘Long first treatment or immediate recurrence’ (6.3%): this group had an average first treatment of 16.9 months (SD = 9.1, median = 19) and nearly half (44.8%) experienced a second episode, which occurred 3.9 months apart (SD = 7.2, median = 3) on average. Their most common last discharge status was completed treatment (42.7%) and late discontinuation (34.9%). (v) ‘Recurrent and decreasing’ (14.2%): this group underwent two (68.2%), three (22.8%), or four or more (9.0%) treatments, with the longest first treatment among the recurrent classes with an average of 8.9 months (SD = 7.5, median = 7) and subsequent shorter treatments. This group had the highest percentage of early discontinuation as their last discharge status (20.7%) among the recurrent classes. (vi) ‘Early discontinuation with recurrence’ (9.9%): this group experienced mostly two (45.7%) treatment episodes, although 26.8% experienced four or more and spent 5.0 months (SD = 4.1, median = 4) in their first treatment on average. Just 11.7% completed treatment and a quarter (26.1%) had an early discontinuation discharge status. Approximately a quarter (25.4%) ultimately were discharged with completed treatment status. Finally, the (vii) ‘Recurrent after long between-treatment periods’ (5.7%): this group had two treatment episodes in most cases (55.9%), but spent 63.2 months (or 5.3 years) on average (SD = 31.1, median = 74) outside the system between their first and second attempt. They present the highest percentage of patients in treatment at the administrative censorship date (16.0%).

FIGURE 1.

Trajectories of substance use treatment utilization since first admission, seven class solution (n = 6266). Note: Class 1 = early discontinuation (32%); class 2 = less than a year in treatment (19.7%); class 3 = year-long episode, without recurrence (12.3%); class 4 = long first treatment, or immediate recurrence (6.3%); class 5 = recurrent and decreasing (14.2%); class 6 = early discontinuation with recurrence (9.9%); class 7 = recurrent after long between treatments period (5.7%).

Factors associated with the treatment trajectory groups

We conducted a multinomial logistic regression model to assess factors associated with the groups using the early discontinuation (class 1) as the reference group (see Table 3). Geographically, participating in treatments from the south and/or the north (vs center) macrozones of Chile at first entry increased the odds of belonging to long one-episode trajectory groups as well as those characterized by recurrence. Having received more intensive treatment modalities at first entry (inpatient or outpatient high intensity vs outpatient low intensity) increased the odds of belonging to the longer one-episode trajectory groups, as well as some trajectories with recurrence when compared to the reference. For instance, receiving care in inpatient facilities (vs outpatient low intensity) increased the odds of belonging to the long first treatment or immediate recurrence (OR = 1.88; 95% CI = 1.30, 2.73) and early discontinuation with recurrence (OR = 2.17; 95% CI = 1.59, 2.97) groups, compared to the early discontinuation group. Regarding ownership of the treatment center at first entry, being treated in private centers decreased the odds of belonging to groups with long one-episode trajectories groups. For instance, it decreased the odds of belonging to the less than a year in treatment (OR = 0.72; 95% CI = 0.59, 0.88), and the long first treatment or immediate recurrence (OR = 0.63; 95% CI = 0.47, 0.86) trajectories when compared to the reference group.

TABLE 3.

Multinomial logistic regression model for the treatment utilization trajectories.

	Less than a year in treatment (19.7%) vs early discontinuation (32.0%)		Year-long episode, without recurrence (12.3%) vs early discontinuation (32.0%)		Long first treatment, or immediate recurrence (6.3%) vs early discontinuation (32.0%)		Recurrent and decreasing (14.2%) vs early discontinuation (32.0%)		Early discontinuation with recurrence (9.9%) vs early discontinuation (32.0%)		Recurrent after long between treatments period (5.7%) vs early discontinuation (32.0%)
	aOR	95% CI	aOR	95% CI	aOR	95% CI	aOR	95% CI	aOR	95% CI	aOR	95% CI
Geographical macrozone of the treatment center (Ref. center)
North	1.23	(0.97, 1.55)	1.63	(1.26, 2.12)	1.33	(0.94, 1.88)	1.35	(1.06, 1.72)	0.97	(0.72, 1.30)	1.69	(1.21, 2.37)
South	1.22	(0.86, 1.74)	1.79	(1.24, 2.57)	1.75	(1.12, 2.73)	1.61	(1.09, 2.38)	1.2	(0.73, 1.96)	2.17	(1.30, 3.64)
Type of treatment at entry (Ref. outpatient low intensity)
Outpatient high intensity	1	(0.84, 1.19)	1.08	(0.88, 1.33)	1.48	(1.14, 1.92)	1.3	(1.07, 1.59)	1.61	(1.28, 2.01)	1.22	(0.92, 1.60)
Inpatient	1.51	(1.18, 1.92)	1.28	(0.96, 1.72)	1.88	(1.30, 2.73)	1.84	(1.40, 2.42)	2.17	(1.59, 2.97)	1.27	(0.85, 1.90)
Ownership of the treatment center at first entry (Ref. public)
Private	0.72	(0.59, 0.88)	0.81	(0.64, 1.02)	0.63	(0.47, 0.86)	0.85	(0.68, 1.06)	0.76	(0.59, 0.98)	0.77	(0.56, 1.05)
Urbanicity of the commune of residence (Ref. rural)
Mixed	1.99	(1.19, 3.33)	1.63	(0.97, 2.76)	0.67	(0.34, 1.32)	1.77	(1.01, 3.12)	0.82	(0.41, 1.65)	1.94	(0.85, 4.39)
Urban	1.59	(1.05, 2.42)	1.11	(0.73, 1.69)	0.73	(0.45, 1.16)	1.55	(0.98, 2.45)	1.23	(0.74, 2.06)	1.67	(0.86, 3.25)
Sex (Ref. men)
Women	0.96	(0.79, 1.16)	1.03	(0.83, 1.29)	1.4	(1.07, 1.84)	1.26	(1.03, 1.54)	1.46	(1.16, 1.83)	1.36	(1.02, 1.82)
Age group (Ref. 18–29)
30–39	1.3	(1.08, 1.56)	1.54	(1.23, 1.91)	1.57	(1.18, 2.10)	1.1	(0.90, 1.34)	1.09	(0.87, 1.36)	1.28	(0.97, 1.69)
40–49	1.1	(0.87, 1.39)	1.8	(1.37, 2.35)	1.75	(1.22, 2.49)	1.12	(0.86, 1.45)	0.91	(0.67, 1.23)	1.2	(0.82, 1.74)
50 or more	1.39	(1.00, 1.93)	2.31	(1.60, 3.33)	2.7	(1.73, 4.24)	0.77	(0.51, 1.18)	0.53	(0.30, 0.91)	0.84	(0.45, 1.59)
Education (Ref. more than high school)
Completed high school or less	1.22	(1.03, 1.45)	1.11	(0.91, 1.36)	1.09	(0.84, 1.41)	1.11	(0.92, 1.34)	1.09	(0.88, 1.35)	1.15	(0.88, 1.51)
Completed primary school or less	1.65	(1.31, 2.08)	1.53	(1.17, 1.99)	1.18	(0.82, 1.68)	1.03	(0.78, 1.35)	0.9	(0.65, 1.24)	1.12	(0.76, 1.65)
Employment (Ref. unemployed)
Employed	1.15	(0.97, 1.37)	1.08	(0.88, 1.31)	1.33	(1.02, 1.73)	1.08	(0.89, 1.31)	0.99	(0.80, 1.24)	0.98	(0.75, 1.29)
Inactive	1.22	(0.94, 1.57)	1.08	(0.80, 1.46)	1.18	(0.81, 1.71)	1.18	(0.90, 1.55)	1.34	(0.99, 1.80)	1.14	(0.77, 1.68)
Marital status (Ref. married/shared living arrangements)
Separated/divorce	1.11	(0.87, 1.42)	0.92	(0.69, 1.22)	0.9	(0.62, 1.31)	1.12	(0.85, 1.48)	1.15	(0.83, 1.58)	0.99	(0.65, 1.50)
Single	0.86	(0.71, 1.04)	0.96	(0.77, 1.19)	1.14	(0.86, 1.52)	0.9	(0.73, 1.10)	0.91	(0.72, 1.15)	0.96	(0.72, 1.29)
Widower	0.79	(0.34, 1.82)	0.79	(0.32, 1.97)	1.21	(0.47, 3.08)	1.24	(0.55, 2.80)	0.37	(0.08, 1.63)	1.35	(0.43, 4.17)
Cohabitation (Ref. family)
Others	0.85	(0.66, 1.11)	1.05	(0.79, 1.42)	0.87	(0.58, 1.31)	0.79	(0.59, 1.07)	0.78	(0.55, 1.10)	0.5	(0.30, 0.83)
Alone	0.79	(0.60, 1.04)	0.9	(0.66, 1.22)	0.61	(0.39, 0.95)	1.06	(0.79, 1.41)	0.88	(0.61, 1.27)	0.56	(0.34, 0.94)
No. of children (Ref. none)
One	0.93	(0.75, 1.14)	1	(0.78, 1.27)	0.78	(0.56, 1.09)	1.03	(0.82, 1.30)	1.23	(0.94, 1.60)	0.79	(0.57, 1.09)
Two	0.95	(0.75, 1.20)	0.89	(0.68, 1.18)	0.88	(0.61, 1.26)	1.06	(0.82, 1.37)	1.24	(0.92, 1.66)	0.86	(0.60, 1.23)
Three or more	0.73	(0.57, 0.95)	0.72	(0.53, 0.97)	0.95	(0.65, 1.39)	0.74	(0.55, 0.98)	0.82	(0.59, 1.15)	0.62	(0.41, 0.93)
Primary substance (Ref. alcohol)
Cocaine	0.84	(0.67, 1.07)	0.99	(0.75, 1.30)	0.91	(0.63, 1.31)	1.38	(1.03, 1.86)	1.49	(1.07, 2.07)	1.32	(0.89, 1.96)
Cocaine paste	0.65	(0.53, 0.80)	0.67	(0.53, 0.85)	0.73	(0.54, 0.99)	1.4	(1.10, 1.80)	1.29	(0.97, 1.72)	1.14	(0.81, 1.60)
Cannabis	0.91	(0.69, 1.22)	1.29	(0.94, 1.76)	0.94	(0.60, 1.45)	1.36	(0.96, 1.93)	1.18	(0.78, 1.79)	0.73	(0.43, 1.26)
Other	1.15	(0.70, 1.90)	1.47	(0.87, 2.49)	2.23	(1.27, 3.93)	2.25	(1.30, 3.88)	0.82	(0.35, 1.94)	0.78	(0.29, 2.08)

Note: Log-likelihood = −10 995. AIC = 21756.01. n = 6157. Unimputed.

Abbreviation: aOR, adjusted odds ratio.

Being a woman was associated with greater odds in all the recurrent trajectories groups. Age showed consistent patterns where belonging to older age groups (vs 18–29 years old) was associated with longer trajectories of one treatment. Less education was associated with belonging to trajectories with more time in treatment and no recurrence when compared to the early discontinuation group. Being employed (vs unemployed) increased the odds of belonging to the long first treatment or immediate recurrence (OR = 1.33; 95% CI = 1.02, 1.73), and being economically inactive (vs unemployed) increased the odds of belonging to the early discontinuation with recurrence group (OR = 1.34; 95% CI = 0.99, 1.80; P = 0.054), relative to the early discontinuation group. Declaring cocaine paste (vs alcohol) as primary substance was the only substance that decreased the odds of belonging to long one-episode trajectory groups, in addition, it increased the odds of belonging to the recurrent and decreasing (OR = 1.4; 95% CI = 1.10, 1.80) and the early discontinuation with recurrence (OR = 1.29; 95% CI = 0.97, 1.72; P = 0.08) groups, relative to the early discontinuation group. People who primarily used cocaine (vs alcohol) shared the same pattern regarding recurrence, as it increased the odds of belonging to the recurrent and decreasing (OR = 1.38; 95% CI = 1.03, 1.86), and the early discontinuation with recurrence (OR = 1.49; 95% CI = 1.07, 2.07) groups, when compared to the early discontinuation group.

In the sensitivity analysis 89.6% of participants were assigned to the same class when using half the time points and the size and direction of the associations presented in the main analysis remained stable (Tables S10–S12).

DISCUSSION

In this study, we used LCGA to characterize SUDs treatment trajectories over 9 years using a rich registry-based retrospective cohort on a national level in the understudied context of Chile. Our study identified seven distinctive treatment utilization groups. Specifically, we identified three groups (classes 1–3) that were not characterized by recurring episodes and had different treatment lengths, a mixed group (class 4) that had a long first treatment or two treatment episodes with a short between-treatment episodes period, as well as three recurrent treatment groups (classes 5–7). Prior studies conducted in high-income countries show that SUDs treatment readmission is fairly common [6, 12, 19–22]. In our cohort, we found 1.5 treatment episodes on average during the 9 years of observations, with 34.2% of individuals having more than one episode, therefore, showing less overall usage of services over time.

SUD patients who repeatedly seek treatment do so in distinct ways, which cannot be captured when dichotomizing the population into first-timers and people with previous treatment [20, 29–31]. LCGA allowed us to identify three recurrent groups showing different treatment use patterns, with shared and unique associated factors. Patients showing persistent recurrent SUD treatment use are of special interest as they exhibit high levels of treatment use with potentially low levels of improvement [1, 23]. In our study, ~10% of the cohort showed persistent contact with the system over the 9 years (class 6). Interestingly, this group was more likely to be out of the labor force and have <50 years, which in the case of Chile might relate to a permanent disability. People who repeat treatment tend to have a worse prognosis and a greater perceived need for services in areas other than substance use [12, 41]. The focus on continuum care [13, 54, 55], long-term recovery management [56] or the provision of other types of services (i.e. peer-based support groups or harm reduction services) in addition to treatment might benefit these patients. Harm reduction, which took off in part as a public health response because of the HIV/AIDS epidemic in other regions [57, 58], is scarce in Chile and Latin America outside of the official treatment system, which could be explained by how low opioid and injection drug use prevalences are [37]. A harm reduction approach acknowledges that not all patients will respond or seek treatment, concentrating efforts on tackling the harms associated with substance use [59]. The provision of pipes and other paraphernalia to people who smoke cocaine paste [60] or managed alcohol programs, which are being evaluated in the global north [61], are not even discussed as public health responses. Harm reduction is also needed among people who inject drugs (PWID), as Latin America has the highest prevalence of HIV among PWID (37.5% vs 17.8% globally) [62]. The acute care model for SUD continues to be the norm. In Chile, although governmental guidelines invite treatment centers to maintain continuity of care with their patients after discharge, those services are currently not financed, threatening their feasibility.

Moreover, we were able to identify different groups with just one episode of care, but distinct times in treatment and associated characteristics (e.g. discharge status). For a better public health response, those in class 1 are of special interest, as they discontinued treatment early and never returned (32%). Previous studies among people who use cocaine found that some patients responded immediately to treatment, staying short times and in later long-term remission. However, those who did not respond and continued with high levels of cocaine use remained out of treatment over the 12-year follow-up [14]. Although we do not know the specific reasons behind their discontinuation, which can be highly diverse [63], it is important to recognize early signs of non-response and improve re-engagement interventions or referral to other services [23], especially among those that discontinue because of reasons associated with their substance use or related risk factors. Evidence shows that very few of those with SUDs receive treatment [64]. At the same time, treatment is the most common intervention employed by governments to tackle SUDs. Therefore, to increase treatment’s impact properly designed and implemented early re-engagement and retention strategies are critical, because they increase time in treatment, which has been consistently associated with better long-term outcomes [6, 14, 21, 24].

Regarding correlates to the trajectory groups, our results highlight the potential influence of certain treatment characteristics on treatment utilization patterns and discontinuation over time [3, 42]. Being a woman was consistently associated with recurrent groups [6, 19, 29, 40]. Women suffer from greater social stigma than men because of their substance use, are a minority in the treatment system and face additional barriers to access and continue treatment, including caregiving [65]. Although in Chile there are women-only outpatient and inpatient programs they are not available for most [46]. Moreover, the risk of readmission between women-only and mixed-gender programs is similar [66]. The design of treatment policies and guidelines for re-engagement and continuity of care should explicitly address these gender differences [65]. Qualitative studies could help understand and explain the recurrence in treatment shown by women. Last, patients with a more severe SUD are more likely to seek treatment repeatedly [20, 29, 30, 39]. In our case, patients who primarily used cocaine (powder or paste) had a greater treatment recurrence, with cocaine paste use also associated with the early discontinuation group compared to those who declared alcohol. A recent meta-analysis of in-person psychosocial treatments highlighted that heavier cocaine use was associated with higher discontinuation [67], which might explain the large percentage of people engaged in treatment for a short period in our cohort. Efforts should be placed among those with cocaine paste use, who constitute almost a third of the adult population treated by the Service in Chile [68]. Contingency management has shown to be successful in reducing cocaine use and increasing retention [69], but it is not used due to, in part, economic, moral and philosophical barriers [70].

This study has several limitations. First, we only had information for treatment episodes funded by SENDA. Therefore, we were unable to capture treatment episodes within SENDA’s network, but funded by the Ministry of Health or out-of-pocket. This likely reduced the number of recurrences, as some participants might have entered treatment funded by other means. However, SENDA funds ~80% of publicly funded treatments in Chile [43], therefore, capturing the majority of episodes. Second, we have no information regarding important covariates and parallel processes, such as a concurrent longitudinal assessment of substance use [23], victimization, psychiatric comorbidities and utilization of other acute care services or peer-based support groups. We also lack information on mortality a competing process that can alter the groups [71]. Likewise, we lacked more detailed information on substance use at first entry (e.g. polysubstance use), geographical and neighborhood-level characteristics, which are influential for premature discontinuation [42]. Third, we explore the trajectories only descriptively. Fourth, the long follow-up may affect treatment seekers in 2023 differently than those in 2010, however, national representative surveys show stability in most prevalences of drug use in Chile between 2010 and 2020 [72]. Last, although LCGA managed to characterize the patterns of treatment utilization over time, it did not provide as much information regarding the type of treatment or discharge status over time, as there were even people in the early discontinuation group who had completed treatment. This highlights the complexity when examining and interpreting SUDs treatment utilization patterns over time using administrative data [22]. Further research could explore other longitudinal techniques such as sequence analysis, which has been recently used to study care trajectories among other health conditions [73–77]. Although very useful, governmental administrative datasets usually hide several services provided to those with SUDs, like peer-group support or harm reduction initiatives, which constitute an important part of their care trajectories and related outcomes. Moreover, terms like ‘completed treatment’ reflect how embedded the acute care model is in our culture and public health response, although SUDs might be reoccurring for some patients.

CONCLUSIONS

Our study provides novel information regarding heterogeneity in SUDs treatment utilization over time in the under-studied context of a middle-income country in Latin America. This exploratory analysis highlights the heterogeneous trajectories of SUD treatment that naturally occur across the population, helping to identify groups with particularly high utilization of services over time and those who discontinue treatment prematurely. In Latin America, a call for harm reduction, continuity in care and evidence-based approaches to treating cocaine use disorder are especially necessary.

DATA AVAILABILITY STATEMENT

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.