Results of the search

The updated database searches for this update produced the following results: MEDLINE 224 (78 new references), Embase 190 (92 new), AMED 6 (1 new), CINAHL 11 (2 new), Cochrane Neuromuscular Specialised Register 66 (33 new), and CENTRAL 63 (37 new). We reviewed 242 new references from database searches for this update, with 217 after deduplication, but none were eligible for inclusion. We also reviewed 77 records in ClinicalTrials.gov and 2 records in WHO ICTRP. See Figure 1 for a flow chart of the study selection process.

Figure 1

---

---

A flow diagram illustrating the study selection process.

We also searched the Database of Abstracts of Reviews of Effects (4 references), Health Technology Assessment database (4 references), and NHS Economic Evaluation Database (3 references) for information for inclusion in the Discussion.

Included studies

All review authors agreed on the inclusion of two studies (Jackson 2001; Bourke 2006).

The first study was a prospective randomised three‐month study in three ALS centres in the USA (Jackson 2001). The study included 20 people with ALS (no age or sex provided), of whom 13 were randomised when overnight oximetry studies documented oxygen saturation below 90% for at least one cumulative minute throughout the duration of the study (a minimum of six hours) and the individual had at least two significant symptoms of nocturnal hypoventilation. There were two groups: an early group where the participants were started immediately on NIV (seven participants) and a late group (six participants) in whom NIV was initiated when forced vital capacity (FVC) was less than 50% predicted. The report provided no demographic characteristics for the participants. The effects of NIV on ALSFRS respiratory version, Pulmonary Symptom Scale, and SF‐36 were estimated. No survival data were available from this study.

The second study was an RCT performed in a single ALS centre in the UK. It included 92 participants, of whom 41 met the criteria for randomisation (orthopnoea or maximum inspiratory pressures less than 60% or symptomatic hypercapnia, or both) and were followed up for at least 12 months or until death (Bourke 2006). Random allocation was computer generated by minimisation, a process that allows all significant prognostic factors to be included in the model. Twenty‐two participants were assigned to the NIV group and 19 participants to the standard care group. Demographic and functional characteristics of the participants in the two groups were similar at randomisation (Characteristics of included studies). The effect of NIV on survival and quality of life outcome domains (e.g. SF‐36 and the Sleep Apnea Quality of Life Index (SAQLI)) was estimated in the whole cohort but also in the subgroups of participants with and without severe impairment of bulbar function. Trialists defined severe impairment as a score of three or less on a simple six‐point assessment scale.

Excluded studies

We excluded one study that was terminated early due to problems recruiting participants (Perez 2003). We excluded Pinto 2003 as it used historical controls. One study was a controlled study of exercise in people with ALS with respiratory insufficiency (Pinto 1999). One study was an anecdotal study (Sivak 1982). Twelve studies were retrospective (Goulon 1989; Kamimoto 1989; Bach 1993; Cazzolli 1996; David 1997; Buhr‐Schinner 1999; Kleopa 1999; Saito 1999; Cedarbaum 2001; Winterholler 2001; Lo Coco 2007; Shoesmith 2007), of which only Kleopa 1999 had a control group.  Six studies were non‐randomised prospective observational studies (Pinto 1995; Aboussouan 1997; Lyall 2001b; Newsom‐Davis 2001; Lo Coco 2006; Mustfa 2006). Berlowitz 2016 was a cohort study examining the effects of NIV on pulmonary function decline and survival across MND/ALS phenotypes.

Pinto 1995 was a single‐centre study in Portugal that included 20 consecutive participants with bulbar features and probable or definite ALS according to El Escorial criteria. The first 10 participants were treated with oxygen, bronchodilators, and physiotherapy. The following 10 participants were submitted to bilevel positive airway pressure. Two participants were excluded, one in the first group because the participant had tracheostomy, and the other in the second group for refusing the treatment. Despite the limitation of small sample size, the study demonstrated a significant improvement in total survival time (P < 0.004) and in the survival from the onset of diurnal gas exchange disorder (P < 0.006). Although it was a controlled trial, participants were not randomised, and hence selection bias cannot be excluded.

A 2016 study looked at the early initiation of NIV within a sham and an interventional arm (Jacobs 2016). It essentially compared 4 cm continuous positive airway pressure against bilevel ventilation at 8/4 cmH₂O at a stage at which NIV would not normally be used. The two groups used NIV (or sham) for only 2.0 hours/day and 3.3 hours/day, respectively. The reported outcome was decline in percentage predicted FVC during the follow‐up period. Once both groups reached the time when they met the standard criteria for starting NIV, both groups were offered NIV. There was no difference in the survival of the two groups. The review authors considered that the main objective of this study was to assess the benefits of early initiation of NIV, and the trial did not evaluate NIV against no intervention, therefore we could not consider this data as complementary to the objectives of this review.

We excluded four apparently unpublished trials in clinical trials registries that compared different settings, modes, or timings of NIV (NCT00537446; NCT00560287; NCT01363882; NCT01641965). We excluded an ongoing trial in which different modes of ventilation were to be compared (NCT01746381). We enquired about two further registry entries from investigators. The contact person for NCT00958048 informed us that because participants were not willing to be randomised, investigators converted the trial to an observational study, which was not eligible for this review. An investigator of NCT00386464 reported that the trial had recruitment problems and was neither completed nor published.

Allocation

Allocation was adequately concealed in both included studies. Bourke 2006 performed centralised randomisation; random allocation was computer generated by the process of minimisation. Although the Jackson 2001 paper gives no data for allocation concealment, the study authors informed us that an independent statistician performed randomisation by preparing two sets of random assignment in blocks of four, which were then allocated to each centre.

Blinding

Blinding of participants was not possible in the included studies, as it was not possible to blind delivery of NIV. Outcome assessors were blinded in Jackson 2001; no information was given in Bourke 2006 as to whether outcome assessors were blinded to knowledge of allocation intervention.

Incomplete outcome data

Jackson 2001 provided data on only six participants from the early‐intervention group. The outcome of one randomised participant was not clear, and the paper provided no data for the late‐intervention group. We noted no obvious attrition bias in Bourke 2006. Thirteen participants withdrew during surveillance, but none withdrew after randomisation, and all participants were followed up until the end of the study or death.

Selective reporting

Bourke 2006 was free of the suggestion of selective outcome reporting, as the study protocol was available, and the trial authors reported all primary and secondary outcomes in the prespecified way. No study protocol was available for Jackson 2001. The results of Jackson 2001 were intended to be used as preliminary data, but unfortunately the investigators did not secure funding for a subsequent study.

Other potential sources of bias

In Jackson 2001, trial authors did not define nocturnal hypoventilation according to the universally accepted criteria; they accepted oxygen desaturation below 90% for at least one cumulative minute as evidence of nocturnal hypoventilation. Nocturnal oximetry showing oxygen desaturation below 88% for at least five consecutive minutes or oxygen saturation below 90% for more than 5% of sleep time are considered sufficient to initiate NIV (Mehta 2001).

Primary outcome

Bourke 2006 showed that the overall median survival after initiation of assisted ventilation was significantly different between the NIV and standard care groups (P = 0.0062). The median survival for the NIV group participants was 48 days longer than the standard care group participants (219 days compared to 171 days). The published information did not provide a 95% confidence interval (CI) for the median survival difference (48 days) or for our secondary outcomes. We were able to derive approximate CIs from P values and median survival estimates under an assumption that median survival follows a lognormal distribution. A statistical method based on a lognormal survival model gave the following estimates for the 95% CI: 12 to 91 days for the estimated 48‐day survival difference (private communication from statistical referee Dan Moore).

The overall median survival of the subgroup with better (good or moderately impaired) bulbar function was also significantly different in the NIV group (P = 0.0059), with NIV group participants surviving 205 days longer than the standard care group participants (median 216 days in NIV group versus 11 days in the standard care group). 

In participants with poor bulbar function, NIV did not confer survival advantage (P = 0.92), with an overall median survival for NIV group participants of 39 days less than the standard care group (222 versus 261 days). We considered the quality of the survival evidence as moderate, as it was based on a single RCT (see summary of findings Table for the main comparison).

Secondary outcomes