Two independent authors (PS, HB) searched PubMed/MEDLINE and PsycInfo through 15/08/2021 without language restrictions, supplemented by a manual review of reference lists from eligible publications and relevant reviews and meta-analyses. The search terms used were: (“infant crying” OR “crying problem*” OR “excessive crying” OR “persistent crying” OR “feeding problem*” OR “feeding disorder” OR “refusal to eat” OR “picky” OR “choosy” OR “infant sleep” OR “sleeping disorder*” OR “regulation disorder*” OR “regulatory problem*” OR “regulatory disorder*”) and (“attention-deficit” OR “attention deficit” OR hyperactivity OR ADHD OR ADD OR hyperkinetic OR “behavioral problems” OR “behavioral outcome” OR “emotional problem” OR “internalizing” OR “externalizing” OR “dysregulated behavior” OR dysregulation OR anxiety OR fear OR psychopathology OR “clinical symptoms” OR preschool OR “growing up” OR follow-up OR “follow up” OR longitudinal OR prospective OR “mental health” OR “epidemiology”). Authors were contacted for additional information.

Inclusion criteria were: (i) prospective study, (ii) reporting on ≥20 children with (iii) RPs regarding crying, sleeping, and/or eating problems (study-defined, see Table 1) (iv) during infancy (≤18 months of age (41)), and (v) more than one follow-up assessment during childhood (2–14 years of age) (vi) reporting on behavioral problems such as externalizing, internalizing, and/or ADHD symptoms. We did not include studies that only reported RPs as outcome at follow-up.

If studies with a community sample only reported continuous measures of RPs of the complete sample at baseline and did not distinguish between RPs and healthy controls (HCs), they were not included. We also excluded studies restricted to children with any kind of disability, pervasive developmental or autism spectrum disorder or those investigating other clinical outcomes only (e.g., eating disorder, obesity, developmental disorder, neurological outcomes). If studies reported on HCs, these data were used irrespective of whether they were assessed prospectively or retrospectively.

Primary outcomes were (i) the cumulative incidence of overall behavioral problems of infants with any RPs (single and multiple) during infancy and (ii) the overall behavioral problems of infants with any RPs (single and multiple) compared to HCs. The outcome of overall behavioral problems summarized the amount and severity of symptoms for the primary and secondary outcomes using the total behavioral problem scores (such as CBCL total score). If those measures were not available, measures of externalizing (preferred), internalizing, and/or ADHD symptoms were supplemented. Both continuous and categorical outcomes were included (42, 43).

Secondary outcomes included (iii) overall behavioral problems of infants with single RPs compared to HCs, (iv) overall behavioral problems of infants with multiple RPs compared to healthy controls, and (v) overall behavioral problems of infants with multiple RPs compared to those with single RPs.

Other outcomes included the cumulative incidence of overall behavioral problems in infants with any RPs and HCs as well as overall behavioral problems and in-between group differences (as described above) of (vi) externalizing problems, (vii) internalizing problems, and (viii) ADHD symptoms separately. If a study reported multiple measures for one outcome, we chose the scale used most often in the overall study sample to ensure homogeneity (for details see Table 1 and Online Resource 1 Supplementary Table S1).

Data of each study were independently identified and extracted by more than 2 authors (HB; PS); inconsistencies were resolved involving a third reviewer (BG). Unadjusted outcome data were preferred. When more than 2 samples with different symptom severity of the same RPs were studied, we extracted data from the sample with the most severely reported symptoms. If continuous and categorical data were reported, continuous data were preferred. Whenever data were missing, authors were contacted for more information.

In the case of overlapping samples, we included the most suitable data (largest sample size and/or matching inclusion criteria). In this context, we decided in one case (18) to extract outcomes from a limited study sample that better matched the predefined follow-up age (9.5 years instead of 4 years). All outcomes were extracted separately for the following groups: (a) any RPs, (b) single RPs, (c) multiple RPs, and (d) HCs. In the group of any RPs, outcomes of infants with single or multiple RPs were pooled, and the incidences/severity scores were merged.

If multiple follow-up assessments were available, we preferred those at age 5–11 years for the main analysis, as most children were assessed at that age. In case of multiple follow-up assessments between age 5–11 years, the later time point was preferred. Additionally, all other follow-up time points were extracted for a separate analysis of age groups. For the studies with multiple follow-up time points, we extracted and used multiple data from different time points. For the category “definition of RP” a studies definition was considered strict, if the assignment to the RP group was based on a structured interview or questionnaire, and it was considered lenient if it was based on a one-item parent report.

Study quality was evaluated via the Newcastle-Ottawa Scale (NOS), which is a scale used for assessing the quality of nonrandomized studies in meta-analyses in the three categories 1) selection of study groups, 2) comparability of the groups and 3) ascertainment of outcome of interest. Data regarding the study quality of each study were independently identified and extracted by ≥2 authors (HB; PS). The overall result is indicated by the NOS score, where a score of ≥7 out of 9 indicates high study quality (44).

The cumulative incidence was computed as the number of children with mental or behavioral problems at follow-up divided by the total number of individuals in the population at risk.

Between-group differences were described for each outcome where more than 2 studies were available using the standardized mean difference (SMD). The SMD was either extracted directly or calculated from means, standard deviations (SDs) and sample sizes, odds ratios, F-statistics, or correlation coefficients (45). The SMD was adjusted using the small sample size bias correction (Hedges' g) (42). SMDs were considered small if between 0.2–0.49, medium if between 0.50–0.79, and large if ≥0.80 (46). Data were analyzed using R (R Core Team, 2019) and Comprehensive Meta-Analysis Version 3 (Borenstein, 2013). All analyses used a random effects model (47), were two-sided, with alpha = 0.05, and were presented as point estimates and corresponding 95% confidence intervals (CIs). In addition, p-values were used to describe the test-for-null outcome effects. Heterogeneity among studies was assessed with I² value (48), with p < .05 and I²≥ 50% indicating significant heterogeneity. Publication bias was assessed for the primary outcomes by using funnel plots and Egger's regression test for funnel plot asymmetry (42, 49) for analyses with ≥10 studies. In addition, we used the trim and fill method which yields an estimate of the effect size after the publication bias (50) and the fail-safe test (estimated number of studies needed in order to obtain a non-significant result).

The following subgroup analyses were added for all outcomes: (i) sample (community vs. clinical), (ii) RP definition (strict vs. lenient), (iii) type of RPs (crying vs. sleeping vs. eating; only for comparison of single RPs vs. multiple and single RPs vs. HC), and (iv) age at baseline (≤6 vs. > 6 months). We conducted random effects meta-regression analyses to identify potential moderators including (i) age at baseline, (ii) age at follow-up, (iii) percentage male, (iv) sample size, and (v) study quality (NOS). To account for a potential change of symptoms with age, we added an additional analysis for any RPs looking at the overall behavioral problems reported for the following age groups only: 2–6, 7–10, 11–14.

The initial search resulted in 3,794 hits. Altogether, 3,705 studies were excluded on the title/abstract level. Of the remaining 89 references 59 articles were excluded after full text review, yielding 30 studies (10, 12, 13, 15, 17–20, 32, 51–69) (Figure 1) that were included in this meta-analysis.

A total of 30 studies reported on 34,582 participants (RPs: n = 5091, control: n = 29,491; baseline = 6.5 ± 4.5 months, follow-up = 5.5 ± 2.8 years, male = 52%). Single RPs were examined in 25 studies (n = 3897; crying problems: studies = 13, n = 1577; sleeping problems: studies = 9, n = 2014; eating problems: studies = 3, n = 105; not specified: studies = 2, n = 201), multiple RPs in 9 studies (n = 1194). Five studies reported co-occurrence of RPs but only analyzed the outcome of single RPs (11, 13, 56, 60, 62) (Table 1, Online Resource Supplementary Table S1).

The overall study quality was high with a mean NOS score of 6.2 ± 1.3 (median = 6, 95% CI = 6–7) and a NOS ≥7 (indicating high study quality) in 13 of 30 studies (43%) (Online Resource Supplementary Table S2).

The meta-analytically calculated cumulative incidence of overall behavioral problems in children with previous RP was 0.233 (95% CI = 0.179–0.298, studies = 18, n = 2873), indicating that 23.3% of those with RPs during infancy developed behavioral problems later (Table 2). Of note, is that this is nearly 4 times more frequent than in HCs that had a cumulative incidence of overall behavioral problems of 0.067 (95% CI = 0.043–0.104, studies = 10, n = 3699). No significant subgroup differences or moderating effects emerged in children with previous RPs, except for smaller sample sizes that were associated with higher overall behavioral problems (p = 0.022) (Online Resource Supplementary Table S3).

When meta-analytically comparing the overall behavioral problems of children with previous single and multiple RPs during infancy to HCs (studies = 26, n = 31,177), those with RPs had significantly more behavioral problems (SMD = 0.381, 95% CI = 0.296–0.466, p ≤ 0.0001) (Figure 2).

The Egger's test (intercept = 1.422, 95% CI = 0.6–2.25, p = 0.003) indicated potential publication bias. After adjustment for 6 potentially missing studies using the trim-and-fill method, the SMD decreased to 0.339 (95% CI = 0.243–0.434).

No significant effects emerged in the subgroup or meta-regression analyses. However, effects sizes were particularly high (SMD = 0.685, 95% CI = 0.295–1.074, p ≤ 0.001) in the clinical sample compared to the community sample (SMD = 0.348, 95% CI = 0.275–0.422, p ≤ 0.001) (Table 2). Effect sizes were alike, regardless of the follow-up age (see Table 3).

The overall behavioral problems of children with single RPs (studies = 23, n = 26,789) and with multiple RPs during infancy (studies = 8, n = 7474) were significantly more pronounced compared to HCs (single RPs: SMD = 0.372, 95%, CI 0.281–0.462, p ≤ 0.0001; multiple RPs: SMD = 0.419, 95% CI 0.200–0.639, p ≤ 0.001). Children with multiple RPs during infancy did not show significantly more overall behavioral problems than those with single RPs (studies = 4, n = 961, SMD = 0.149, 95% CI = −0.250–0.549, p = 0.463) (see Online Resource Supplementary Table S4).

The overall behavioral problems of children with excessive crying (SMD = 0.493, 95% CI = 0.336–0.651, p ≤ 0.001) and sleeping problems (SMD = 0.266, 95% CI = 0.138–0.395, p ≤ 0.001) were significantly more pronounced compared to HCs, while no significant effect on childhood outcomes emerged in those with eating problems during infancy (SMD = 0.231, 95% CI = −0.021–0.482) (see Table 2 and Online Resource Supplementary Table S3). In-between subgroup differences (crying vs. sleeping vs. eating problems) were not significant (p = 0.128).

Children with previous RPs were also more frequently affected by externalizing problems (RPs: cumulative incidence = 0.201, 95% CI = 0.141–0.279; HCs: cumulative incidence = 0.067, 95% CI = 0.028–0.151, SMD = 0.362, 95% CI = 0.253–0.472, p = 0.001), internalizing problems (RPs: cumulative incidence = 0.160, 95% CI = 0.120–0.209; HCs: cumulative incidence = 0.083, 95% CI = 0.063–0.109, SMD = 0.343, 95% CI = 0.284–0.403, p ≤ 0.001), and ADHD (RPs: cumulative incidence = 0.242, 95% CI = 0.157–0.354; HCs: cumulative incidence = 0.076, 95% CI = 0.026–0.201, SMD = 0.461, 95% CI 0.317–0.605, p = 0.071) (Table 2). For subgroup and meta-regression analyses see Online Resource Supplementary Table S5.

The findings of the conducted meta-analysis are consistent with the existing meta-analysis by Hemmi et al. (16) and extend the scope of the negative impact of RPs on childhood behavioral problems. A considerable effort was made to include as much data as possible in this meta-analysis whilst maintaining strict inclusion criteria: Additional information from eleven studies was included in this meta-analysis. Using strict inclusion criteria and methodological rigor, we aimed to rule out as many sources of potential bias as possible. In this context, we excluded three studies that Hemmi et al. (16) had included (including a control group with transient RPs and a follow-up age younger than two years). Moreover, we expanded methods by also reporting on the cumulative incidence of behavioral problems during childhood.

There was high heterogeneity in the data. Identified studies were heterogeneous with respect to sample characteristics, RP definition, measurement instruments, number of subscales, outcomes, and follow-up duration (range 2.5–11 years) which likely contribute to the heterogeneity of the results. Many different forms of defining and assessing RPs existed across countries and centers. Consequently, the use of standardized tools that focus on parent and infant behaviors for enhanced comparability in further research is needed. Since most studies assessed infant RPs and child behavior using parent reports rather than objective measures or clinician observation, a reporting bias might influence the results. For example, maternal “overrating” of the children's behavior might be rooted in maternal stress and/or the continued perception of the child as being difficult (20, 73). Therefore, in future research more objective and multi-informant measures of child behavior (e.g., clinical observation, teacher reports), parental characteristics, and the parent-infant interaction should be used (86).

Moreover, some studies showed co-occurrence of RPs but only reported behavioral problem outcomes for single RPs and did not control for any other RPs. Previous research found that crying, sleeping, and eating problems often coexist (6, 13). Hence, non-reported or non-assessed co-occurrence of RPs might lead to a biased conclusion regarding the effects of single RPs. More longitudinal prospective studies are needed to enable a profound investigation of the association of RPs, behavioral problems, and potential confounders. Although we included a set of study-specific moderators in the meta-regressions, the inclusion of other essential moderators, such as maternal depression, preterm birth, parent-infant interaction, childhood trauma or childhood attachment was limited because comparable information of potentially relevant confounders was often lacking.

From a clinical perspective, our findings highlight the need for a better understanding of predictors of childhood behavioral problems and clinical disorders. The results suggest the importance of early monitoring, detection, and intervention for families with an infant affected by RPs to prevent the development of further behavioral problems. From a primary health care perspective, this is crucial information for pediatricians and parental counseling in childcare.

Overall, the knowledge about the impact of RPs on later behavioral problems should be used to develop and evaluate specific prevention programs focusing, for example, on parent-infant psychotherapy (87, 88). The mutual perspective of RPs offers several optional starting points for interventions for disrupted parent-infant interactions to reduce parental stress and foster further child development. Strengthening parentś self-efficacy to adapt to their infant´s needs and difficulties is just as important as identifying and treating emotional distress, particularly postpartal depression or anxiety which might keep parents from adequately understanding or supporting their infant (70).