Keith J BarringtonCHU Sainte Justine

MontréalCanada

The systematic reviewUpdate of the Cochrane Review13 RCTs of iNO in preterm infants with

respiratory diseaseVarying eligibility, age of use, indications,

dose, duration.Decided Post Hoc to divide them into 3

groupsEarly rescueLate ‘rescue’Early routine treatment

MortalityNone of the individual trials show a

statistically significant reduction.Meta-analysis of the first 2 trials of early

routine use previously showed potential difference: which was marginally significant.

Potentially important magnitude of effect, but the estimate lacked precision.

Adding the data from the abstract of the EUNO trial: now no difference, RR 0.91 (95% CI 0.74, 1.11)

The early rescue, and later treatment in infants at risk of BPD groups show no effect on mortality.

Survival without bronchopulmonary dysplasiaEarly rescue treatment studies: no apparent

benefitsubgroup analysis of Van Meurs et al reduction in infants > 1 kg birthweight.

Early routine treatment: modest and almost significant reduction in the combined outcome of death or BPD RR 0.93, 95% CI 0.86 1.01.

Heterogeneity I2 = 34%, Kinsella + EUNO: no benefit. Subgroup analysis of Kinsella suggested lower risk infants (BW >1000 g) had a benefit.

In Schreiber, subgroup analysis suggested only the less sick infants (OI less than the median) benefited.

Survival without bronchopulmonary dysplasiaBallard et al report: significant benefit of inhaled

nitric in improving survival without BPD. However the figures given, 165/294 vs. 182/288

are not significant using RevMan to calculate RR.

Difference due to use of multiple outputation as planned, a statistical way of correcting for possible coherence among multiples, (1st of multiples to be eligible randomized, therefore entered as a cluster, should be analyzed as a cluster).Subgroup analysis: reduction in combined death or BPD when started at 7- 14 d, less severe disease may increase chance of benefit.

Brain Injury

Early rescue studies: no significant effect, the direction of the difference is toward increased serious IVH Not significant at p<0.05, BUT potential evidence of harm should always be taken seriously, especially when there is no evidence of benefit.

Later entry based on BPD risk not expected to affect IVH incidence.

With the addition of EUNO 2009, early routine use shows no effect on brain injury, either severe IVH or combined outcome of severe IVH or PVL.

Heterogeneity I2= 80%

Brain InjuryPrevious analysis suggested a potential

benefit in the early routine treatment groupEUNO had a higher baseline OI, but a lower

mortality than the other early routine treatment trials, Do these characteristics explain the difference, or is it just chance?

Sensitivity analysisI have performed a sensitivity analysis of the

early rescue treatment studies, eliminating the studies with very early primary outcomes, the studies with potential crossover, and studies with very small samples

Analyses including Innovo, Kinsella 1999, and Van Meurs 2005, total n=566

Conclusions are substantially unchanged

Problems with the meta-analysisGreat variation in study designDifferent eligibility criteria

Postnatal age severity of illness criteria

Different primary outcomesVariable duration of therapyVariable doses

IPD meta-analysisThe standard review and meta-analysis

suggested that there might be significant benefits in certain groups of infants:

But no clear indications for therapy were evident

Don’t want to miss benefitsNor potential harmsIn some groups of infants

IPD meta-analysisObtaining original information on all the

enrolled subjects Allows:

Intention to treat analysisConsistency of definitions (if enough

information was collected)Analyses of subgroups across trialsIdentification of patient characteristics

associated with positive or negative responses

Meta Analysis of Preterm Patients on inhaled Nitric Oxide Collaboration

Inhaled nitric oxide in preterm infants: a systematic review and individual patient data meta-analysis

• To determine whether inhaled nitric oxide in preterm infants receiving assisted ventilation improves survival without morbidity, specifically without: – CLD– cerebral injury– retinal injury– long-term disability

• To determine whether the effects of inhaled nitric oxide differ according to the risk profile of the patient in terms of:

– gestational age at birth– severity of illness– antenatal steroid use– postnatal age at the time of randomization– ventilation mode at randomization– administration of exogenous surfactant– inhaled nitric dosage and duration of nitric oxide administration

• determine whether certain patient or treatment characteristics may predict benefit from inhaled nitric oxide in the premature infants (patient-level subgroup analyses)

• define CLD the same way across trials

• aim to analyse all endpoints in all patients in all trials, to avoid biases associated with selective publication endpoints

• address additional research questions

• improve trial identification, interpretation & dissemination via collaborative approach

• main analyses used log-binomial regression models adjusted for trial

• subgroup analyses used poisson regression models with robust error variance due to issues with model convergence

• endpoints between siblings from multiple births

• the primary analyses were adjusted for multiple births using the multiple outputation approach (1000 repetitions)

• sensitivity analysis: generalized estimating equations (GEE) used to analyse the two main endpoints of interest

Distribution of multiplesby treatment group

Trial n % multiples iNO Placebo

Van Meurs 2005 22 4.9 10 (45.5%) 12 (54.5%)

Srisuparp 2002 2 5.9 0 (0.0%) 2 ( 100%)

Kinsella 2006 130 16.4 61 (46.9%) 69 (53.1%)

Hascoet 2005 14 9.7 10 (71.4%) 4 (28.6%)

Schreiber 2003 26 12.6 11 (42.3%) 15 (57.7%)

Kinsella 1999 2 2.4 1 (50.0%) 1 (50.0%)

Dani 2006 2 5.0 2 ( 100%) 0 (0.0%)

EUNO 2008 152 19.0 71 (46.7%) 81 (53.3%)

Ballard 2006 84 14.4 47 (56.0%) 37 (44.0%)

OVERALL 434 13.1 213 (49%) 221 (51%)

Main endpoints assessable for all patients (patients with no event were censored)

• Death or chronic lung disease (CLD) using best available definition (alive and O2 dependent at 36 weeks PMA if calculable, or trialists’ own

definition)

• Severe adverse neurological event after randomization (intracranial hemorrhage (IVH) grade III or IV, or cystic periventricular leukomalacia (PVL) or other pathologies such as periventricular echodensity, periventricularcysts, ventriculomegaly or hydrocephalus)

† Subhedar removed from the analysis as zero cell counts caused model instability.

* χ2 test for heterogeneity p > 0.05

Estimates derived from N=1000 iterations of log-binomial model using multiple outputation method.

Death or CLD (Best available definition)

0.2 0.5 1 2 5

OVERALL*

iNO

40 / 48 (83%)

6 / 16 (38%)

43 / 105 (41%)

42 / 61 (69%)

54 / 64 (84%)

170 / 224 (76%)

292 / 398 (73%)

4 / 20 (20%)

165 / 294 (56%)

134 / 399 (34%)

954 / 1629 (59%)

RR (95% CI)Trial Placebo

27 / 32 (84%)

4 / 18 (22%)

56 / 102 (55%)

51 / 84 (61%)

56 / 62 (90%)

174 / 225 (77%)

294 / 395 (74%)

8 / 20 (40%)

184 / 288 (64%)

137 / 401 (34%)

992 / 1627 (61%)

Favours placeboFavours iNO

EUNO 2008

Ballard 2006

Dani 2006

Kinsella 2006

Van Meurs 2005

INNOVO 2005

Hascoet 2005

Schreiber 2003

Srisuparp 2002

Kinsella 1999

0.96 (0.91, 1.01) p=0.095

1.01 (0.83, 1.23)

0.85 (0.74, 0.98)

0.53 (0.19, 1.46)

0.99 (0.91, 1.08)

0.98 (0.88, 1.09)

0.93 (0.82, 1.07)

1.11 (0.85, 1.43)

0.77 (0.57, 1.04)

1.59 (0.55, 4.62)

0.99 (0.81, 1.21)

OVERALL*

RR (95% CI)Trial

Severe Neurological Event

† Data from Ballard not available for this analysis.

* χ2 test for heterogeneity p > 0.05

0.1 0.2 0.5 1 2 5

1 / 20 (5%)

8 / 48 (17%)

1 / 16 (6%)

16 / 105 (15%)

31 / 61 (51%)

5 / 64 (8%)

69 / 224 (31%)

117 / 398 (29%)

8 / 20 (40%)

81 / 399 (20%)

315 / 1649 (19%)

3 / 22 (14%)

7 / 32 (22%)

4 / 18 (22%)

25 / 102 (25%)

28 / 84 (33%)

6 / 62 (10%)

52 / 225 (23%)

112 / 395 (28%)

7 / 20 (35%)

68 / 401 (17%)

286 / 1649 (17%)

Estimates derived from N=1000 iterations of log-binomial model using multiple outputation method.

Favours placeboFavours iNO

1.12 (0.98, 1.29) p=0.089

1.19 (0.88, 1.61)

1.13 (0.50, 2.52)

1.04 (0.83, 1.30)

1.35 (0.99, 1.83)

0.81 (0.26, 2.51)

1.57 (1.06, 2.34)

0.63 (0.35, 1.13)

0.27 (0.03, 2.13)

0.76 (0.31, 1.88)

0.37 (0.04, 3.25)

EUNO

Dani 2006

Kinsella 2006

Van Meurs 2005

INNOVO 2005

Hascoet 2005

Schreiber 2003

Srisuparp 2002

Kinsella 1999

Subhedar 1997

iNO Placebo

† Data on deaths only available from Ballard.

* χ2 test for heterogeneity p=0.050

Estimates derived from N=1000 iterations of log-binomial model using multiple outputation method.

Severe Neurological Event (including death)

RR (95% CI)Trial iNO Placebo

OVERALL*

Ballard 2006

EUNO 2008

Dani 2006

Kinsella 2006

Van Meurs 2005

INNOVO 2005

Hascoet 2005

Schreiber 2003

Srisuparp 2002

Kinsella 1999

Subhedar 1997 11 / 20 (55%)

27 / 48 (56%)

3 / 16 (19%)

27 / 105 (26%)

42 / 61 (69%)

36 / 64 (56%)

143 / 224 (64%)

170 / 398 (43%)

10 / 20 (50%)

110 / 399 (28%)

22 / 294 (7%)

601 / 1649 (36%)

8 / 22 (36%)

22 / 32 (69%)

5 / 18 (28%)

40 / 102 (39%)

36 / 84 (43%)

41 / 62 (66%)

128 / 225 (57%)

171 / 395 (43%)

8 / 20 (40%)

92 / 401 (23%)

21 / 288 (7%)

572 / 1649 (35%)

0.1 0.2 0.5 1 2 5

1.06 (0.98, 1.16) p=0.149

1.03 (0.58, 1.83)

1.22 (0.95, 1.57)

1.25 (0.62, 2.50)

1.01 (0.86, 1.19)

1.13 (0.97, 1.31)

0.85 (0.64, 1.13)

1.61 (1.18, 2.20)

0.66 (0.43, 1.01)

0.64 (0.18, 2.24)

0.82 (0.58, 1.16)

1.51 (0.77, 2.99)

Favours placeboFavours iNO

* χ2 test for heterogeneity p > 0.05

Estimates derived from N=1000 iterations of log-binomial model using multiple outputation method.

OVERALL*

iNO

11 / 20 (55%)

22 / 48 (46%)

2 / 16 (13%)

16 / 105 (15%)

28 / 61 (46%)

33 / 64 (52%)

114 / 224 (51%)

78 / 398 (20%)

4 / 20 (20%)

22 / 294 (7%)

54 / 399 (14%)

384 / 1649 (23%)

Trial

Death at any time

Placebo

7 / 22 (32%)

17 / 32 (53%)

2 / 18 (11%)

22 / 102 (22%)

25 / 84 (30%)

36 / 62 (58%)

102 / 225 (45%)

94 / 395 (24%)

6 / 20 (30%)

21 / 288 (7%)

41 / 401 (10%)

373 / 1649 (23%)

Favours placeboFavours iNO

EUNO 2008

Ballard 2006

Dani 2006

Kinsella 2006

Van Meurs 2005

INNOVO 2005

Hascoet 2005

Schreiber 2003

Srisuparp 2002

Kinsella 1999

Subhedar 1997

0.1 0.2 0.5 1 2 5 10

1.05 (0.93, 1.17) p=0.45

1.41 (0.95, 2.09)

1.03 (0.58, 1.83)

0.70 (0.23, 2.10)

0.85 (0.65, 1.11)

1.13 (0.93, 1.37)

0.89 (0.65, 1.22)

1.53 (0.98, 2.39)

0.72 (0.39, 1.33)

1.06 (0.17, 6.67)

0.86 (0.55, 1.36)

1.73 (0.83, 3.58)

RR (95% CI)

* χ2 test for heterogeneity p > 0.05

Estimates derived from N=1000 iterations of log-binomial model using multiple outputation method.

OVERALL*

iNO

10 / 20 (50%)

22 / 48 (46%)

2 / 16 (13%)

15 / 105 (14%)

28 / 61 (46%)

32 / 64 (50%)

114 / 224 (51%)

78 / 398 (20%)

4 / 20 (20%)

22 / 294 (7%)

54 / 399 (14%)

381 / 1649 (23%)

Trial

Death before discharge

Placebo

7 / 22 (32%)

17 / 32 (53%)

2 / 18 (11%)

20 / 102 (20%)

25 / 84 (30%)

36 / 62 (58%)

97 / 225 (43%)

93 / 395 (24%)

6 / 20 (30%)

21 / 288 (7%)

41 / 401 (10%)

365 / 1649 (22%)

Favours placeboFavours iNO

EUNO 2008

Ballard 2006

Dani 2006

Kinsella 2006

Van Meurs 2005

INNOVO 2005

Hascoet 2005

Schreiber 2003

Srisuparp 2002

Kinsella 1999

Subhedar 1997

0.1 0.2 0.5 1 2 5 10

RR (95% CI)

1.06 (0.94, 1.19) p=0.32

1.41 (0.95, 2.09)

1.03 (0.58, 1.83)

0.70 (0.23, 2.10)

0.86 (0.65, 1.13)

1.19 (0.97, 1.46)

0.86 (0.62, 1.19)

1.53 (0.98, 2.39)

0.75 (0.39, 1.43)

1.06 (0.17, 6.67)

0.86 (0.55, 1.36)

1.57 (0.74, 3.34)

Severe intra-ventricular hemorrhage

Data from Ballard not available for this analysis.

* χ2 test for heterogeneity p > 0.05

Estimates derived from N=1000 iterations of log-binomial model using multiple outputation method.

† Subhedar removed from the analysis as zero cell counts caused model instability.

RR (95% CI)Trial iNO Placebo

Favours iNO0.01

OVERALL*

EUNO 2008

Dani 2006

Kinsella 2006

Van Meurs 2005

INNOVO 2005

Hascoet 2005

Schreiber 2003

Srisuparp 2002

Kinsella 1999 3 / 47 (6%)

1 / 6 (17%)

13 / 105 (12%)

14 / 61 (23%)

2 / 41 (5%)

67 / 189 (35%)

60 / 398 (15%)

3 / 18 (17%)

71 / 399 (18%)

234 / 1264 (19%)

2 / 30 (7%)

3 / 5 (60%)

19 / 102 (19%)

14 / 83 (17%)

3 / 32 (9%)

46 /164 (28%)

74 / 395 (19%)

4 / 19 (21%)

56 / 401 (14%)

221 / 1231 (18%)

0.1 0.2 0.5 1 2 5 10

1.04 (0.88, 1.23) p=0.61

1.27 (0.91, 1.77)

0.84 (0.22, 3.22)

0.81 (0.59, 1.12)

1.26 (0.93, 1.73)

0.52 (0.09, 2.93)

1.39 (0.70, 2.78)

0.67 (0.34, 1.33)

0.28 (0.04, 1.91)

0.95 (0.17, 5.36)

Favours placebo

This analysis assumes that if IVH is missing then IVH = 0 for on study results.

Data from Van Meurs, Srispuarp, Subhedar, Kinsella 99 and Ballard not available for this analysis.

* χ2 test for heterogeneity p > 0.05

Estimates derived from N=1000 iterations of log-binomial model using multiple outputation method.

† Data have been removed where Baseline HUS was taken after start of study gas and also where baseline HUS showed the presence of a Grade 4 IVH

Severe intra-ventricular hemorrhage worse than baseline IVH

RR (95% CI)Trial iNO Placebo

0.2 0.5 1 2 5 10

OVERALL 0.87 (0.68, 1.12) p=0.29

EUNO 2008 1.26 (0.81, 1.94)

Dani 2006 1.23 (0.13, 11.48)

Schreiber 2003 0.37 (0.16, 0.88)

Hascoet 2005 1.02 (0.35, 2.96)

Kinsella 2006 0.78 (0.55, 1.11)

INNOVO 2005 0.65 (0.11, 3.73)2 / 64 (3%) 3 / 62 (5%)

49 / 398 (12%) 63 / 394 (16%)

9 / 51 (18%) 5 / 24 (21%)

6 / 26 (23%) 8 / 13 (62%)

2 / 14 (14%) 1 / 8 (13%)

43 / 300 (14%) 36 / 309 (12%)

111 / 853 (13%) 116 / 810 (14%)

0.1

Data from Van Meurs, Srispuarp, Subhedar, Kinsella 99 and Ballard not available for this analysis.

* χ2 test for heterogeneity p > 0.05

Estimates derived from N=1000 iterations of log-binomial model using multiple outputation method.

† Data have been removed where Baseline HUS was taken after start of study gas

Severe intra-ventricular hemorrhage adjusted for worse baseline HUS result

RR (95% CI)Trial iNO Placebo

OVERALL

EUNO 2008

Dani 2006

Schreiber 2003

Hascoet 2005

Kinsella 2006

INNOVO 2005 1 / 35 (3%) 2 / 28 (7%)

60 / 360 (17%) 74 / 348 (21%)

12 / 53 (23%) 6 / 25 (24%)

6 / 26 (23%) 8 / 13 (62%)

3 / 14 (21%) 1 / 8 (13%)

46 / 305 (15%) 41 / 312 (13%)

128 / 793 (16%) 132 / 734 (18%)

0.01 0.1 0.2 0.5 1 2 5 10

0.87 (0.71, 1.08) p=0.21

1.10 (0.74, 1.63)

2.31 (0.67, 7.95)

0.23 (0.08, 0.62)

1.06 (0.47, 2.42)

0.83 (0.63, 1.09)

0.40 (0.04, 4.01)

Subgroup Analyses(death or CLD)

Overall n (%) 95% CI

Subgroup Group iNO Placebo

Relative

Risk LCL UCL p

P

(heterogeneity)*

Trials with median Start dose >5ppm <=5 ppm 732 / 1194 (61%) 739 / 1199 (62%) 1.00 0.94 1.06 0.889 <0.001

>5 ppm 218 / 435 (50%) 252 / 428 (59%) 0.83 0.74 0.95 0.005

Trials with median Start dose >10ppm <=10 ppm 779 / 1319 (59%) 803 / 1321 (61%) 0.98 0.92 1.04 0.413 N.S.

>10 ppm 171 / 310 (55%) 188 / 306 (61%) 0.87 0.76 0.99 0.041

Exposure to iNO (trial level) † Low 353 / 522 (68%) 372 / 525 (71%) 0.95 0.87 1.04 0.254 N.S.

High 597 / 1107 (54%) 619 / 1102 (56%) 0.96 0.89 1.03 0.216

Exposure to iNO (individual level)‡ <105 ppm*days 247 / 452 (55%) 252 / 467 (54%) 1.03 0.92 1.16 0.564 N.S.

>=105 ppm*days 344 / 639 (54%) 363 / 617 (59%) 0.89 0.82 0.98 0.014

Data from Subhedar removed from these analyses as zero cell counts caused model instability.

†Srisuparp, Kinsella 2006, EUNO 2008 and Ballard classified as high exposure trials, others classified as low exposure. High exposure was defined as area

under the dose-time curve >= 70 calculated on a trial specific basis.

‡Per-patient exposure analysis only included three trials (Kinsella 06, EUNO 2008 and Ballard). All other trials were excluded either because dose exposure

was dependent on response (INNOVO, Van Meurs, Subhedar, Srisuparp, Hascoet and Kinsella99) or because data on length of time on treatment was not

available (Dani, Schreiber). (These data were not initially requested). Exposure was calculated as AUC and patients were grouped as less than the median

(105) or not.

Estimates derived from N=1000 iterations of a poisson regression model with robust error variance using multiple outputation method.* p value for heterogeneity was calculated as the proportion of the 1000 models which had non-significant p values for the interaction term for subgroup*treatment effect (N.S. = non significant)

Overall n (%) 95% CI

Outcome iNO Placebo Relative Risk LCL UCL p value

Death or CLD 956 (59%) 993 (61%) 0.96 0.91 1.00 0.0610

Severe Neurological Event 337 (25%) 312 (23%) 1.11 0.98 1.27 0.1072

Death at any time 386 (23%) 374 (23%) 1.03 0.92 1.16 0.5916

Death by 36 weeks 350 (21%) 336 (20%) 1.05 0.93 1.19 0.4504

Death by discharge 383 (23%) 366 (22%) 1.05 0.93 1.17 0.4481

Severe IVH 234 (19%) 221 (18%) 1.04 0.88 1.23 0.6404

ImplicationsFurther analysis of individual patient data will allow

definition of patient characteristics most promising for future studies,

and most promising treatment protocols, including dose and

duration.Based on a single high quality study, later prolonged

treatment with iNO seems to decrease BPD among at-risk infants, without adverse effects,

The benefit of early routine treatment which appeared possible after 2 trials were completed now seems less likely after EUNO, further analysis of the IPD may help to clarify.

Unanswered questionsPreterm infants with clear evidence of PPHN

exist, is it reasonable to withhold iNO based on these analyses?

There were 142 babies with ‘pulmonary hypertension’ in the database, with a small non-significant improvement in the primary outcome

Diagnostic criteria for PPHN in these infants are uncertain

Infants with acute deterioration after the early neonatal period, improvements in oxygenation may follow iNO, such babies were not eligible for the majority of these studies.Does iNO improve their survival or long term

outcomes?

Implications for further researchMost promising appears to be infants at high

risk for BPD who are still intubated at 7 days of age

Treatment should be prolonged and commence at more than 5 ppm