Health trajectories of patients with amyotrophic lateral sclerosis before and after initiation of non-invasive ventilation: a French nationwide database analysis Authors: Pierre Tankéré, Emilien Bernard, Eleonore Herquelot, Hélène Denis, Lilian Sfeir, Christel Saint-Raymond, Martial Mallaret, Florent Lavergne, Sébastien Baillieul, Laure Peter-Derex, Renaud Tamisier, Jean Louis Pépin Published: 2026 ABSTRACT AbstractBackgroundManagement of amyotrophic lateral sclerosis (ALS) is complicated by heterogeneous presentation and unpredictable disease course. This study described disease trajectories before and after initiation of non-invasive ventilation (NIV) therapy in individuals with ALS, examined the relationship between NIV initiation timing and survival and analysed health trajectory clusters.MethodsData were extracted from the French national health insurance reimbursement system database for individuals with ≥1 reimbursement for NIV from January 2015 to December 2019, and ≥1 ALS disease code. Health trajectory clusters were determined using time sequence analysis through K-clustering.ResultsWe analysed data from 3443 individuals with ALS (58% male, median age 67 years). The median (IQR) time from ALS diagnosis to NIV initiation was 10.8 (4.5–22.2) months and death occurred 21.5 (12.8–33.9) months after diagnosis. Tracheostomy/gastrostomy was performed in 3.9%/33.4% of patients, respectively. Unsupervised machine learning clustering identified four distinct patient groups. NIV initiation was late in two Clusters (A and B); these individuals were younger, had fewer comorbidities and more physiotherapy sessions before/after NIV. Survival after NIV initiation was longer in Clusters B and C; these individuals had lower rates of depression/anxiety, more prescription of mechanical in/exsufflation therapy and fewer home and emergency hospitalisations. Cluster B was unique, showing late NIV initiation and long post-NIV survival. This cluster was more likely to have spinal onset, a higher rate of obstructive sleep apnoea and fewer comorbidities.ConclusionsThere was marked heterogeneity between patients with ALS and their care trajectories. Our data do not support a universal benefit for early initiation of NIV therapy. WHAT IS ALREADY KNOWN ON THIS TOPIC Patients living with amyotrophic lateral sclerosis (ALS) have heterogeneous disease trajectories. Treatment with non-invasive ventilation (NIV) in patients with ALS has been reported to improve quality of life and mortality. However, the optimal timing for NIV initiation is still debated. Patients living with amyotrophic lateral sclerosis (ALS) have heterogeneous disease trajectories. Treatment with non-invasive ventilation (NIV) in patients with ALS has been reported to improve quality of life and mortality. However, the optimal timing for NIV initiation is still debated. WHAT THIS STUDY ADDS Using a nationwide exhaustive French claims database, we identified four distinct clusters of different pre-NIV and post-NIV health trajectories in people with ALS, showing that heterogeneity extends beyond bulbar/spinal patterns and that early NIV does not universally improve survival in this population. Using a nationwide exhaustive French claims database, we identified four distinct clusters of different pre-NIV and post-NIV health trajectories in people with ALS, showing that heterogeneity extends beyond bulbar/spinal patterns and that early NIV does not universally improve survival in this population. HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY Knowledge gained from data-driven identification of distinct care trajectories in NIV-treated patients with ALS might help to better define individualised timing of NIV based on patient profiles and care trajectories. Our data could be used to inform guidelines and design of future trials. Knowledge gained from data-driven identification of distinct care trajectories in NIV-treated patients with ALS might help to better define individualised timing of NIV based on patient profiles and care trajectories. Our data could be used to inform guidelines and design of future trials. Introduction Amyotrophic lateral sclerosis (ALS) is a neurodegenerative disease of the central nervous system that is ultimately fatal. The most common cause of disability, hospitalisation and death in patients with ALS is respiratory failure, which requires specialised management.1 ALS management is particularly challenging due to the wide variability in clinical presentation.1–3 Health trajectories of patients living with ALS also vary markedly,3 and there is substantial diversity in care organisation.1 The occurrence of milestone medical events (eg, riluzole prescription, gastrotomy, tracheotomy, use of non-invasive ventilation (NIV)) and treatment decisions/sequences of care change the trajectory of management and prognosis for individuals with ALS.3 Current criteria for starting NIV therapy include the symptom burden and findings of pulmonary function tests and blood gas.4–6 The use of NIV in people living with ALS has been shown to improve both survival and quality of life.7 8 However, survival benefits appear less in those with the bulbar-onset ALS phenotype.8 In a randomised trial, early versus late initiation of NIV therapy has been shown to attenuate the decline of respiratory function without improving survival.9 In contrast, a potential survival benefit of early NIV therapy initiation has been reported in retrospective studies.10 11 There is therefore a need to better define respiratory impairment trajectories in people with ALS and determine the most appropriate time in the disease course to initiate NIV therapy to maximise the benefits obtained.8 Machine learning (ie, innovation in visualisation and data analysis) might help in deciphering the heterogeneity in ALS health trajectories.3 Thus, this study aimed to describe pre-NIV and post-NIV disease trajectories in a comprehensive national population of individuals with ALS (bulbar-onset and/or spinal-onset phenotype), determine the relationship between the timing of NIV therapy initiation and survival, and describe health trajectory clusters in these individuals. Methods Study design Data for this observational study were extracted from the French national health insurance reimbursement system database (SNDS).12 It covers 99% of the French population and collects individual pseudonymised data from beneficiaries. This analysis used data from the general insurance scheme (private sector employees), covering approximately 76% of the French population (40 million adults). Unemployed and retired individuals are covered by the scheme relating to their previous occupation. The SNDS contains socio-demographic data (age, sex and area of residence), and details of all healthcare expenses (outpatient visits, reimbursed drug therapy, medical procedures, hospital admission diagnoses), and date of death. In accordance with French regulations, individual patient consent was not necessary since there was no direct contact with individuals. Study population Eligible individuals were those aged ≥25 years with ≥1 reimbursement for NIV between 1 January 2015 and 31 December 2019, and at least one hospital stay or long-term chronic disease registered with the International Classification of Diseases, 10th Revision (ICD-10) code G12.2 (motor neuron disease). The age threshold of ≥25 years was applied to exclude juvenile forms of ALS, which are rare and have a distinct genetic background and specific clinical characteristics.13 Only incident cases (ie, those with no NIV treatment in the previous 5 years) were included. Individuals were followed from NIV initiation until death or the end of the study period (December 2020), whichever came first. Individuals still alive at the study end of follow-up were censored. Outcomes and definitions The primary aim was to describe care trajectories for people with ALS treated with NIV to characterise the phenotypic heterogeneity, intensity of care, key milestones in disease trajectories, timing of NIV initiation, end-of-life support and mortality of this population. A secondary aim was to compare characteristics and follow-up for different health trajectory clusters in the study population. The index date was the date of first delivery of NIV during the study inclusion period. The date of ALS diagnosis was defined as the first occurrence of any of the following: hospitalisation or long-term chronic disease coding with the G12.2 ICD-10 code, dispensing of riluzole, or last electromyogram within the 3 months before prescription of riluzole. Individuals were classified as having bulbar-onset ALS if one of the following occurred within the first 6 months after the ALS diagnosis: hospital stay for dysphagia, swallowing disorder, aspiration pneumonia, aspiration with bronchial obstruction or reimbursement for atropine. Individuals without these characteristics were designated as having ‘spinal-onset ALS’, although in practice this group corresponds to non-bulbar-onset phenotypes, encompassing a heterogeneous set of clinical presentations. Claims in the 5 years before NIV initiation were used to identify comorbidities (based on ICD-10 codes), medical history and to confirm incident status. Baseline demographic variables included age and sex. Clinical comorbidities were identified using ICD-10 diagnostic codes and reimbursement data, and included obesity (E66), diabetes (E10–E14) and obstructive sleep apnoea (OSA; G47.3). Depression, anxiety and other psychiatric disorders were defined based on their respective ICD-10 codes (F32–F41) and/or reimbursement for related pharmacological treatments (see online supplemental material for full details). For OSA, only a diagnosis or continuous positive airway pressure (CPAP) prescription recorded before the ALS diagnosis was considered to ensure that OSA was established independently and before ALS onset. The following events and outcomes were evaluated: time between diagnosis and NIV initiation, tracheotomy, gastrostomy, mechanical in/exsufflation initiation, sleep study, riluzole therapy, NIV usage >12 hour/day and prescription of devices showing loss of ability to walk; and the number of hospitalisations, emergency department visits, medical consultations, physiotherapist consultations, blood gas analysis, pulmonary function testing, nocturnal oximetry, palliative care and death. Codes used to identify each outcome are provided in the online supplemental material. Statistical analysis Data are reported by the classical classification of ALS (bulbar-onset and spinal-onset) and by trajectories clustering supported by unsupervised machine learning analyses. Data on the study population are presented using descriptive statistics (median (IQR) for quantitative variables, and frequency (%) for qualitative variables). Median (IQR) values are estimated based on the number of annualised events per patient, considering only patients with at least one event. Annualised event counts are obtained by dividing the total number of events during the follow-up period by the individual follow-up duration. This approach ensures comparability between patients, irrespective of the completeness of annual data. Bubble plots were generated to represent the median time between ALS diagnosis and the first occurrence of each event of interest. Between-group differences were tested using the Student’s t-test for normally distributed continuous variables, the Wilcoxon rank-sum test for non-normally distributed continuous variables, the χ2 test for categorical variables with frequencies ≥10 in all categories, and Fisher’s exact test for other categorical variables. Normality of distribution for continuous variables was tested using the Kolmogorov-Smirnov test. Machine learning, time sequence analysis through K clustering (TAK analysis) was used to identify clusters of participants who had similar health trajectories, as described previously.14 Main identified variables for identification of clusters were: time between ALS diagnosis and NIV initiation; time between NIV initiation and the end of follow-up; and how follow-up ended (patient death or end of study). To achieve patient representation and clustering, the algorithm was based on a hierarchical agglomerative clustering ward method.15 All analyses were performed using Python V.3.10 and SAS V.9.4 Results Study population A total of 3443 adults living with ALS who were started on NIV were included (58% male, median age at diagnosis 67 years) (table 1). There was a low cardiometabolic comorbidity burden, but nearly half of all participants had depression/anxiety (table 1). The majority of patients had spinal onset; those with bulbar onset were significantly more likely to be female and have a higher comorbidity burden (table 1). There was a high intensity of care before NIV initiation (table 1). Table 1Study population characteristics at baseline and intensity of care before the initiation of non-invasive ventilationOverall (n=3443)Type of ALSBulbar-onset (n=1020)Spinal-onset (n=2423)P valueMale sex, n (%)2003 (58.2)507 (49.7)1496 (61.7)<0.001*Age at diagnosis, years67.0 (60.0– 74.0)69.0 (63.0–76.0)67.0 (59.0–73.0)<0.001†Comorbidities, n (%) Diabetes390 (11.3)135 (13.2)255 (10.5)0.022* Psychiatric diseases/depression/anxiety1650 (47.9)538 (52.8)1112 (45.9)<0.001* Cardiovascular disease877 (25.5)296 (29.0)581 (24.0)0.002* Obesity386 (11.2)130 (12.8)256 (10.6)0.064* Obstructive sleep apnoea608 (17.7)156 (15.3)452 (18.7)0.018*Charlson Comorbidity Score, n (%) 01713 (49.8)444 (43.5)1269 (52.4)<0.001‡ 1–21627 (47.3)540 (52.9)1087 (44.9)<0.001‡ ≥3103 (3.0)36 (3.5)67 (2.7)<0.001‡Intensity of care in the year before NIV initiation Annual number of ambulatory medical consultations9.5 (6.0–14.6)10.9 (6.8–16.2)9.0 (5.8–14.0)<0.001‡ Annual number of physiotherapist consultations67.4 (38.6–93.7)57.5 (29.8–85.7)70.4 (42.2–96.7)<0.001† Annual number of planned or ambulatory hospitalisations2.2 (1.0–4.7)3.5 (1.8–7.0)1.8 (0.8–3.8)<0.001† Annual number of blood gas analysis tests2.2 (1.2–3.7)3.0 (1.8–5.2)1.9 (1.1–3.2)<0.001† Annual number of pulmonary function tests2.2 (1.2–3.7)2.9 (1.5–4.9)2.0 (1.1–3.3)<0.001† Sleep study, n (%)866.0 (25.2)238 (23.3)628 (25.9)0.120*Values are median (IQR) or number of patients (%).*Χ2 test.†Wilcoxon Mann-Whitney rank-sum test.‡Fisher’s exact test.ALS, amyotrophic lateral sclerosis.NIV, non-invasive ventilation. Study population characteristics at baseline and intensity of care before the initiation of non-invasive ventilation Values are median (IQR) or number of patients (%). Χ2 test. Wilcoxon Mann-Whitney rank-sum test. Fisher’s exact test. ALS, amyotrophic lateral sclerosis.NIV, non-invasive ventilation. Median (IQR) time from ALS diagnosis to NIV initiation was 10.8 (4.5–22.2) months overall, but was shorter in the bulbar-onset (6.1 (2.7–12.1)) versus spinal-onset (13.6 (6.3–26.1)) subgroup (p<0.001, Wilcoxon Mann-Whitney test). Median (IQR) duration of follow-up after NIV initiation was 12.0 (5.3–21.4) months overall, and 9.7 (4.1–17.7) and 13.0 (6.0–23.1) months, respectively, in the bulbar-onset and spinal-onset subgroups (p<0.001, Wilcoxon Mann-Whitney test). Milestones and outcomes During follow-up, more than one-quarter of patients had NIV device usage of >12 hours/day; this proportion was significantly higher in the spinal-onset subgroup but NIV usage ≥12 hours/day occurred significantly earlier in the bulbar-onset subgroup (table 2). Need for mechanical in/exsufflation was also higher and the time to tracheostomy was significantly shorter in the bulbar-onset versus spinal-onset subgroup (table 2). Almost all patients (93%) started treatment with riluzole, three-quarters had mobility loss and one-third required gastrostomy; riluzole use and mobility loss were more common, and gastrostomy less common, in the spinal-onset versus bulbar-onset subgroup (table 2, figure 1). More than half of the population had unplanned hospitalisations (figure 1). Bubble chart showing interventions and support for participants with spinal-onset or bulbar-onset amyotrophic lateral sclerosis. The size of each bubble is proportional to the size of the sample (reference is NIV initiation). ALS, amyotrophic lateral sclerosis; ICU, intensive care unit; NIV, non-invasive ventilation. Table 2Milestones and outcomes for the whole population and in the bulbar-onset versus spinal-onset amyotrophic lateral sclerosis subgroupsOverall(n=3443)Type of ALSBulbar-onset (n=1020)Spinal-onset (n=2423)P valueRespiratory careNIV usage >12 hours/day Patients, n (%)940 (27.3)219 (21.5)721 (29.8)<0.001* Time since diagnosis, months17.2 (9.2–30.6)9.5 (4.5–15.8)21.1 (11.3–35.0)<0.001†In/exsufflation initiation Patients, n (%)1817 (52.8)579 (56.8)1238 (51.1)0.024* Time since diagnosis, months13.5 (6.7–25.5)7.62 (3.9–13.7)17.7 (9.6–30.6)<0.001†Tracheotomy Patients, n (%)135 (3.9)38 (3.7)97 (4.0)0.774* Time since diagnosis, months21.1 (10.5– 40.8)12.0 (5.1–20.5)29.2 (13.5–47.0)0.001†Other medical milestonesRiluzole Patients, n (%)3202 (93.0)942 (92.4)2260 (93.3)0.372* Time since diagnosis, months0.9 (0.2–2.0)0.8 (0.2–1.8)0.99 (0.2–2.1)0.007†Mobility loss Patients, n (%)2551 (74.1)659 (64.6)1892 (78.1)<0.001* Time since diagnosis, months10.4 (4.9–18.9)7.20 (3.3–13.5)11.8 (5.8–20.7)<0.001†Gastrostomy Patients, n (%)1149 (33.4)491 (48.1)658 (27.2)<0.001* Time since diagnosis, months14.5 (7.8–24.3)9.4 (4.9–14.8)19.6 (12.0–31.0)<0.001†Unplanned hospital careRespiratory hospitalisation Patients, n (%)1942 (56.4)591 (57.9)1351 (55.8%)0.253* From NIV initiation, annual number of events2.0 (1.0–3.4)2.4 (1.3–4.0)1.8 (0.93.2)<0.001†ICU hospitalisation Patients, n (%)1447 (42.0)445 (43.6)1002 (41.4)0.232* From NIV initiation, annual number of events1.3 (0.7–2.7)1.7 (0.9–3.4)1.2 (0.7–2.5)<0.001†End-of-life support and mortalityHome hospitalisation Patients, n (%)980 (28.5)294 (28.8)686 (28.3)0.793* Time since diagnosis, months18.8 (10.3– 31.0)11.7 (5.0–18.4)22.6 (13.3–36.2)<0.001†Palliative care Patients, n (%)1787 (51.9)561 (55.0)1226 (50.6)0.020* Time since diagnosis, months17.2 (8.7–28.3)10.7 (4.8–18.6)20.7 (11.6–33.8)<0.001†Death Patients, n (%)2732 (79.3)860 (84.3)1872 (77.3)<0.001* Time since diagnosis, months21.5 (12.8– 33.9)15.5 (9.2–24.1)24.6 (15.5–39.2)<0.001† From NIV initiation, annual number of events8.8 (3.8–16.1)7.2 (3.2–13.5)9.3 (4.3–17.0)<0.001†Values are median (IQR) or number of patients (%).*Χ2 test.†Wilcoxon Mann-Whitney rank-sum test.ALS, amyotrophic lateral sclerosis; ICU, intensive care unit; NIV, non-invasive ventilation. Milestones and outcomes for the whole population and in the bulbar-onset versus spinal-onset amyotrophic lateral sclerosis subgroups Values are median (IQR) or number of patients (%). Χ2 test. Wilcoxon Mann-Whitney rank-sum test. ALS, amyotrophic lateral sclerosis; ICU, intensive care unit; NIV, non-invasive ventilation. Home hospitalisation was not common, but just over half of all patients had palliative care (figure 1); both were needed sooner in the bulbar-onset versus spinal-onset subgroup (table 2, figure 1). The mortality rate was higher (figure 1) and time to death shorter (table 2) in the bulbar-onset subgroup. Characteristics and outcomes by cluster Four different clusters were identified using unsupervised machine learning TAK methodology (table 3, figure 2). Cluster A (n=721; 21%) had the smallest proportion of males and individuals with obesity, and the lowest rate of comorbidities. In this cluster, there was a long time from diagnosis to NIV initiation, but a short time between NIV initiation and death; time from diagnosis to death was in the expected range for patients with ALS. Cluster B (n=765; 26%) was the youngest, had a long time between diagnosis and NIV initiation, NIV initiation and death and diagnosis and death; time from diagnosis to death was longer than the median expected for individuals with ALS. Cluster C (n=844; 25%) had the highest proportion of males (61%) and the highest comorbidity burden. The median time from diagnosis to NIV initiation was short, but the time from NIV initiation to death was long. Cluster D (n=1113; 32%) included individuals with rapidly progressing disease. Median times from diagnosis to NIV initiation, NIV initiation to death and diagnosis to death were all short. Trajectories before and after at-home non-invasive ventilation: visualisation by unsupervised machine learning clustering (TAK) The vertical dashed line in the centre of the figure marks the time of NIV initiation. Each horizontal line represents one patient, with colours indicating events of interest (ALS diagnosis, hospitalisations, NIV usage, palliative care, death; see legend). Clusters A to D represent groups of patients with distinct patterns of care trajectories before and after NIV initiation as defined by unsupervised machine learning clustering. This visualisation highlights the heterogeneity of care pathways across the four clusters. ALS, amyotrophic lateral sclerosis; ICU, intensive care unit; NIV, non-invasive ventilation; TAK, time-sequence analysis through K-clustering. Table 3Characteristics and outcomes by clusterCluster A (n=721)Cluster B (n=765)Cluster C (n=844)Cluster D (n=1113)P valueCharacteristicsMale sex, n (%)390 (54.1)459 (60.0)515 (61.0)639 (57.4)0.280*Age at diagnosis, years66.0 (59.0–73.0)64.0 (55.0–71.0)69.0 (63.0–75.0)70.0 (63.0–76.0)<0.001†Bulbar-onset/spinal-onset ALS, %19.3/80.712.0/88.033.2/66.845.7/54.3<0.001*Comorbidities, n (%) Diabetes50 (6.9)83 (10.9)113 (13.4)144 (12.9)0.001* Depression/anxiety/psychiatric treatments400 (55.5)361 (47.2)345 (40.9)544 (48.9)<0.001* Cardiovascular disease141 (19.6)155 (20.3)231 (27.4)350 (31.5)<0.001* Obesity57 (7.9)85 (11.1)118 (14.0)126 (11.3)0.0024* Obstructive sleep apnoea107 (14.8)185 (24.2)175 (20.7)141 (12.7)<0.001* Charlson Comorbidity Score 0410 (56.9)422 (55.2)368 (43.6)513 (46.1)<0.001‡ 1–2296 (41.1)328 (42.9)446 (52.8)557 (50.0)<0.001‡ ≥315 (2.1)15 (2.0)30 (3.6)43 (3.9)<0.001‡Outcomes Time from ALS diagnosis to NIV initiation, months19.5 (14.7–26.1)31.2 (23.2–46.3)5.5 (2.2–10.0)5.0 (2.3–7.9)<0.00† Death Patients, n (%)720 (99.9)402 (52.5)536 (63.4)1075 (96.6)<0.001* Time from ALS diagnosis, months25.8 (20.6–32.0)51.8 (42.6–68.4)26.3 (21.0–33.4)11.2 (7.7–15.0)<0.001† Time from NIV initiation, months4.8 (2.3–8.2)19.7 (14.3– 26.6)19.7 (15.2–25.8)5.5 (2.6–9.1)<0.001†Values are median (IQR) or number of patients (%).*Χ2 test.†Kruskal-Wallis test.‡Fisher’s exact test.ALS, amyotrophic lateral sclerosis; NIV, non-invasive ventilation. Characteristics and outcomes by cluster Values are median (IQR) or number of patients (%). Χ2 test. Kruskal-Wallis test. Fisher’s exact test. ALS, amyotrophic lateral sclerosis; NIV, non-invasive ventilation. Milestones and outcomes by cluster, and cluster groups Almost all metrics differed significantly between clusters (table 4, figure 2). NIV was started late in Clusters A and B (A-B) and early in Clusters C and D (C-D). Those with late NIV initiation were younger, more likely to have spinal onset, had fewer comorbidities and were less likely to have obesity or diabetes. The A-B group had more physiotherapy sessions before and after NIV, and a slightly higher rate of gastrostomy. Those in Clusters C-D had more respiratory hospitalisations and intensive care unit (ICU) admissions than those in Clusters A-B, and palliative care and home hospitalisation occurred earlier in Clusters C-D versus A-B. Table 4Annual care, follow-up, milestones and outcomes after NIV initiation by clusterCluster A (n=721)Cluster B (n=765)Cluster C (n=844)Cluster D (n=1113)P valueFollow-up Medical ambulatory consultation, number of events12.6 (7.2–21.0)7.0 (4.0–11.2)8.2 (5.3–12.6)12.8 (7.2–21.9)<0.001* Annual number of physiotherapist consultations89.9 (60.1–139.1)93.6 (61.8–133.7)77.6 (47.9– 112.1)83.8 (47.6–127.7)<0.001* Annual number of planned or day hospitalisations3.6 (2.1–5.7)1.2 (0.7– 2.2)1.2 (0.7– 2.4)3.3 (1.9–5.7)<0.001* Annual number of blood gas analysis tests3.3 (2.1–5.7)1.3 (0.8–2.4)1.7 (1.0–2.7)3.3 (1.8–7.0)<0.001* Annual number of pulmonary function tests1.5 (1.2–4.5)0.5 (0.3–0.6)0.6 (0.3–0.8)3.4 (1.2–7.9)<0.001* Patients with a sleep study, n (%)33 (4.6)102 (13.3)99 (11.7)63 (5.7)<0.001†Respiratory careNIV usage >12 hours/day, n (%)151 (20.9)216 (28.2)280 (33.2)293 (26.3)<0.001†In/exsufflation initiation, n (%)377 (52.3)465 (60.8)478 (56.6)497 (44.7)<0.001†Tracheotomy, n (%)13 (1.8)50 (6.5)39 (4.6)33 (3.0)<0.001†Other medical milestones since diagnosis, n (%) Mobility loss542 (75.2)656 (85.8)642 (76.1)711 (63.9)<0.001† Riluzole706 (97.9)704 (92.0)768 (91.0)1024 (92.0)<0.001† Mobility loss542 (75.2)656 (85.8)642 (76.1)711 (63.9)<0.001† Gastrostomy290 (40.2)256 (33.5)271 (32.1)332 (29.8)<0.001†Hospital care and end-of-life careRespiratory hospitalisation From diagnosis, n (%)258 (35.8)368 (48.1)429 (50.8)399 (35.8)<0.001† From NIV initiation, annual number of events3.3 (2.1–6.1)1.3 (0.7–2.2)1.5 (0.8–2.6)2.8 (1.6–4.9)<0.001*ICU hospitalisation From diagnosis, n (%)241 (33.4)254 (33.2)295 (35.0)358 (32.2)0.640† From NIV initiation, annual number of events2.6 (1.6–5.3)0.8 (0.4– 1.2)0.7 (0.5–1.0)2.3 (1.4–4.3)<0.001*Home hospitalisation232 (32.2)189 (24.7)225 (26.7)334 (30.0)0.005†Palliative care (pre or post NIV)436 (60.5)322 (42.1)416 (49.3)613 (55.1)<0.001†Values are median (IQR) or number of patients (%).*Kruskal-Wallis test.†Χ2 test.ALS, amyotrophic lateral sclerosis; ICU, intensive care unit; NIV, non-invasive ventilation. Annual care, follow-up, milestones and outcomes after NIV initiation by cluster Values are median (IQR) or number of patients (%). Kruskal-Wallis test. Χ2 test. ALS, amyotrophic lateral sclerosis; ICU, intensive care unit; NIV, non-invasive ventilation. Survival after NIV initiation was long in Clusters B-C and short in Clusters A-D. Comparisons of B-C versus A-D showed more males, more OSA, less depression/anxiety, fewer medical consultations before and after NIV initiation, less blood gas analysis/pulmonary function testing, more tracheotomy and in/exsufflation and less palliative care, home hospitalisation, emergency hospitalisation and ICU admission. Cluster B had late NIV initiation and longer post-NIV initiation survival than the other clusters. There was a higher proportion of patients with spinal onset, a low comorbidity burden and a higher rate of OSA in Cluster B versus other clusters. The rate of depression/psychiatric treatments was lower in Cluster B versus Cluster A-D, and the proportion of obesity or diabetes was higher than in Cluster D and lower than in Cluster A. Cluster B had the highest annual use of physiotherapy before and after NIV initiation, but the lowest number of medical consultations, planned hospitalisations and blood gas analysis/pulmonary function testing. The proportion of people undergoing a sleep study was highest in Cluster B, but was relatively low in all clusters. Cluster B also had the lowest palliative care and home hospitalisation rates. The rate of respiratory hospitalisations and ICU admissions was lower for clusters with a long survival duration after NIV initiation. Discussion This study highlights the diversity of clinical characteristics in individuals with ALS who require NIV therapy, and the variety of health trajectories before and after NIV initiation, medical milestones including respiratory care and tracheostomy rate on a nationwide exhaustive population as well as end-of-life support. These differences were seen within the traditional bulbar-onset/spinal-onset subgroups used to phenotype patients with ALS, necessitating more advanced analysis to better understand these differences. Therefore, we used unsupervised machine learning TAK analysis to create four clusters of patients whereby each cluster was characterised by similar care trajectories, particularly in terms of NIV initiation delay after diagnosis and survival after NIV initiation. These provided original information about characteristics, intensity of care, medical milestones and outcomes, facilitating better understanding of the determinants of clinical decisions around NIV therapy initiation in ALS and the associated outcomes. The demographics of our ALS cohort started on NIV are consistent with those of large ALS cohorts described in previous publications.1 16 The median time of nearly a year between ALS diagnosis and NIV initiation, and time intervals between diagnosis or NIV initiation and death are also consistent with previous publications.7 17–19 The real-life trajectories of care (intensity of medical and physiotherapist care, respiratory investigations, hospitalisations and end-of-life support) varied substantially between individuals and between the identified four clusters. Rates of use of pulmonary function testing and arterial blood gas analysis were consistent with existing recommendations for respiratory follow-up,6 but the utilisation of sleep study was very low given international guidelines5 6 and the known prevalence of sleep disorders in people with ALS.20 A novel observation from our cluster analysis is that survival time after NIV did not follow a clear pattern based on the timing of NIV initiation. Both early and late NIV initiation were observed in the patient groups that showed short or long survival duration, suggesting that the timing of NIV initiation by itself is not a definitive marker for the clinical trajectory in an individual patient, including speed of respiratory function decline and prognosis. Comparing the grouped cohorts with late versus early initiation of NIV (A-V vs C-D), it appears that starting NIV therapy is primarily driven by the patient profile, trajectories of care and the speed at which respiratory parameters deteriorate. Initiation of NIV was later in spinal-onset and younger patients, in those with fewer comorbidities and in those with a higher number of physiotherapist sessions. European guidelines recommend that the indication for NIV therapy should largely be based on pulmonary function parameters and nocturnal oximetry.6 NIV might also be started during or after acute episodes of respiratory failure.21 Existing French data from a study of 624 patients suggest that the most common indication for NIV therapy in ALS was symptomatic daytime hypercapnia (ie, relatively late in the ALS course).22 However, some of the participants in our study did have early NIV therapy initiation. Current literature emphasises poor tolerability and device usage as risks with the late initiation of NIV therapy.22 23 However, starting NIV early in patients without significant symptoms could also result in low NIV usage.11 21 It is also important to take into account patient factors when deciding to start NIV therapy.24 Looking at our clusters that had a long versus short survival time after NIV therapy initiation (B-C vs A-D), the higher intensity of medical consultations, planned hospitalisations and drug therapy before and after the start of NIV in those with a short time to death suggests that a lack of medical attention is not a likely contributor to poor prognosis in these individuals.25 People with ALS who survived for longer after starting NIV therapy were slightly more likely to be male, although differences in sex distribution between clusters were not statistically significant. Therefore, our data may represent a descriptive trend that has not been observed in previous studies.26 Less depression and anxiety was also associated with a better prognosis in our population. Premorbid anxiety/depression has been reported as a risk factor for shorter survival time in patients with cancer27 but not yet in those with ALS. The higher rate of tracheotomy in clusters that had longer post-NIV initiation survival may be a result of there being more time for the need to develop and a decision to be made to perform this intervention. A decision to go ahead with tracheotomy is something that remains a difficult one from a variety of perspectives (ethical, personal and organisational). Another finding of our study was that the clusters that had longer survival time on NIV had lower use of palliative care and home hospitalisation, and fewer ICU/emergency hospitalisations. This suggests that the patients in these clusters were experiencing slower disease progression and had less progression to end-of-life support. Cluster B in our analysis was characterised by late initiation of NIV and long post-NIV initiation survival. These individuals had predominantly spinal onset of disease, a low comorbidity burden and the highest rate of OSA. The over-representation of individuals with OSA in this cohort is inconsistent with previous data showing a worse prognosis in people with ALS who also have OSA28 and the reported risk of persistent obstructive events during NIV in these patients, resulting in a poor prognosis.29 Possible explanations for these discrepancies are that late NIV initiation could reflect delayed respiratory decline and that previous exposure to CPAP treatment for OSA could improve tolerance of later NIV therapy. Longer survival and late initiation of NIV in our study were associated with the most intensive physiotherapist follow-up, both before and after the start of NIV therapy and in/exsufflation. A recent meta-analysis concluded that physical therapy interventions may improve global function, but that the small number of studies limits the ability to draw definitive conclusions.30 In addition, cough impairment has been shown to be associated with mortality in people with ALS.31 Our findings suggest that the combined use of NIV to compensate for impaired inspiratory muscle function and physiotherapy plus mechanical in/exsufflation to overcome expiratory deficiencies might be a good approach. Key strengths of this study include the size of the cohort, use of the SNDS database (providing a comprehensive national dataset), the duration of follow-up and the novel advanced analytics used to define and describe care trajectories. The main limitations of the study relate to the use of a database that was designed for administrative, rather than research, purposes. Therefore, key data that are missing include details of NIV settings, adherence and efficacy and both neurological and respiratory clinical and patient-reported measures (eg, Revised Amyotrophic Lateral Sclerosis Functional Rating Scale, Center for Neurologic Study Bulbar Function Scale, forced vital capacity, maximum inspiratory and expiratory pressures, peak cough flow, smoking status). In addition, the granularity of the SNDS does not allow detailed phenotypic classification and we applied the classical bulbar-onset versus spinal-onset dichotomy which is widely used in ALS research,32 although more refined classifications have been proposed including progressive muscular atrophy or respiratory onset among others.1 However, the SNDS is a large, unbiased claims database that covers the majority of the French population across healthcare providers, insurers and devices. Moreover, our data are only appropriate to generalise to a similar clinical population of individuals living with ALS who have been started on NIV based on the judgement of their treating clinician. It is also important to remember that patients with ALS who did not have an indication for NIV, or who had an indication for NIV but declined therapy, were not included in the dataset. Furthermore, survival analyses were based on death only, and tracheostomy was examined separately. Although this choice may theoretically introduce bias, <4% of patients underwent tracheostomy, reducing the impact of this clinical milestone on our overall mortality findings. Finally, our definition of early versus late NIV therapy initiation was not based on conventional physiological thresholds such as the forced vital capacity or sniff nasal inspiratory pressure but instead on clusters derived from unsupervised machine learning applied to real-world care trajectories. The data reported reflect actual clinical practice at national level in France and offer a complementary perspective regarding NIV therapy initiation timing. This study documents the marked heterogeneity present across all domains associated with the care of patients with ALS, including patient profiles, the intensity and trajectories of care, frequency of examinations, occurrence of key medical milestones, the timing of NIV therapy initiation, end-of-life support and survival time. Our data highlight the value of physiotherapy and mechanical in/exsufflation as key components of quality care for people living with ALS. Conversely, our data do not support a universal benefit for early initiation of NIV therapy, with some patients who had late NIV initiation showing longer time to death. Some of the novel findings of our study, such as the higher prevalence of OSA in those who showed longer survival after NIV initiation, need to be investigated further. Overall, our findings suggest that the initiation of NIV in people with ALS cannot be guided by any general rule but instead needs to be individualised according to the profile of each patient and the rate of respiratory deterioration (as evidenced by hospitalisations for acute respiratory failure, among other things). Supplementary material