We thank Dr. Elizabeth A Pritts, Dr. David L Olive, and Dr. William H. Parker for their great interest in our study and for the very qualified comments. To investigate the influence of uterine fibroids on obstetrical outcomes is complex and can be viewed from different angles. - We aimed to investigate symptomatic uterine fibroids. Well conducted cohort studies are the best available substitutes when experiments are not possible. Methodological limitations need to be taken into consideration, and associations can be explored. We are fully aware that limitations must be relevantly discussed. David A Grimes 1 pointed out some important issues regarding cohort studies:
1) The exposure should be clear: we argue to have a clearly defined exposure.
2) The importance of potential confounders: we have been able to include important confounders using the directed acyclic graph, DAG.
3) Awareness that the risk of a weak association (HR 0.5-2) is likely to be due to bias and not causation: We have clearly stated that we have found associations and further that the association could be interesting for further research, but we do not recommend changes in clinical recommendations.
The risks of misclassification of exposure were related to the uterine fibroid diagnoses codes. The daily clinical coding may be incorrect or lacking due to various work-related distractions or variable individual interpretation of clinical cases. The low prevalence of uterine fibroids in our study population is likely to be a result of underreporting, or due to DNBC representing a selected group of women.
Specific diagnosis codes within the DNPR have been validated with different results. As you correctly point out, several studies have found specific data within the DNPR invalid, but others have in contrast found other data valid and suitable for research 2-7. A detailed validation of the uterine fibroid diagnosis codes was not possible and has not previously been performed. We relied on previous studies that have shown that reproductive gynecological coding in the DNPR is generally valid and suitable for clinical quality control 8. Further, we made it clear that we investigated a specific group of women with symptomatic uterine fibroids and not uterine fibroids in general. We cannot completely eliminate the risk that some women got the diagnosis code without symptoms, but we found it unlikely, considering that women in Denmark are mainly examined by ultrasound scan due to symptoms, not a routine basis. If the uterine fibroids were symptomatic and required treatment, we found it likely that the women were given a diagnosis code linked to the treatment code. On these ground we argue that we have investigated a group of women with symptomatic uterine fibroids. Others have defined the exposure based on routine ultrasound scans made in the first trimester of pregnancy, irrespective of symptoms. Symptomatic uterine fibroids might represent another group of fibroids with other associated obstetrical risks.
Participants in cohort studies are often a selected group due to the acceptance of participation and may not fully represent the source population. The DNBC mainly consists of white women with middle or high social status 9. We are therefore evaluated a selected group of women, which could explain the lower prevalence of uterine fibroid diagnosis codes.
A potential bias will lead towards exposed women being categorized as unexposed. Since the potential underreporting is independent of the outcome because of the prospective nature of data collection in a cohort study, the potential misclassification could have led to non-differential information bias and hence attenuation of the association between exposure (uterine fibroids) and outcomes 10.
Regarding the very relevant comment on our use of robust standard errors we would like to clarify our decision. Some women contributed with more than one pregnancy in the analyses. Hence, we had clustered data, since observations from two different pregnancies in the same woman would tend to be more similar to each other than to individuals in the rest of the sample. By the use of robust standard errors, we corrected the standard errors for the clustering 11, a well-known and appropriate statistical method.
Response to the comment; "The authors state that risk of preterm birth is decreased, and cesarean section rates increased among women with untreated fibroids compared with women with surgical excision. However, this appears to be based upon raw numbers and not statistical analysis, as the two groups (fibroids in situ versus fibroids removed) were never compared"
In the article we wrote that ‘’ In the present study, the risk of preterm birth decreased whereas the risk of CS increased after myomectomy compared with the risks among women with untreated uterine fibroids‘’ We might have added ”insignificantly” to make it clear that none of the formal comparisons between groups 2 and 4 were significant. However, this is evident from the published odds ratios and 95% CIs in table 2. For example, 1.81 (1.24 – 2.65) is clearly not significantly different from 2.27 (1.30 – 3.96).
Future research on uterine fibroids and impact on obstetrical outcome can be approached in different ways. It could include all uterine fibroids as one big group. Uterine fibroids could be identified and number, type and size of the fibroids recorded. Another approach could be to examine women with clinical symptoms of the uterine fibroids. It is possible that these groups of fibroids are different in terms of associated obstetrical outcomes and thereof potentially different in terms of clinical recommendations.
1.Grimes DA. Epidemiologic research with administrative databases: red herrings, false alarms and pseudo-epidemics. Hum Reprod 2015;30(8):1749-52.) 2.Hess DT Jr. The Danish National Patient Register. Surg Obes Relat Dis. 2016 Feb;12(2):304. doi: 10.1016/j.soard.2015.11.001. Epub 2015 Nov 5. 3.Adelborg K, Sundbøll J, Munch T, Frøslev T, Sørensen HT, Bøtker HE, Schmidt M. Positive predictive value of cardiac examination, procedure and surgery codes in the Danish National Patient Registry: a population-based validation study. BMJ Open. 2016 Dec 9;6(12):e012817. doi: 10.1136/bmjopen-2016-012817.
4.Lund H, Vyberg M, Eriksen HH, Grove A, Jensen AØ, Sunde L.Hydatidiform mole: validity of the registration in the Danish National Patient Registry, the Danish Cancer Registry, and the Danish Pathology Registry 1999-2009. Clin Epidemiol. 2018 Sep 18;10:1223-1231. doi: 10.2147/CLEP.S169657. eCollection 2018. 5.Loft ND, Andersen CH, Halling-Overgaard AS, Thyssen JP, Skov L, Egeberg A.Validation of Psoriasis Diagnoses in the Danish National Patient Register. Acta Derm Venereol. 2019 Oct 1;99(11):1037-1038. doi: 10.2340/00015555-3278. 6.Guillain-Barré syndrome in Denmark: validation of diagnostic codes and a population-based nationwide study of the incidence in a 30-year period. Clin Epidemiol. 2019 Apr 18;11:275-283. doi: 10.2147/CLEP.S199839. eCollection 2019. 7.Kirkegård J, Mortensen MR, Johannsen IR, Mortensen FV, Cronin-Fenton D.Positive predictive value of acute and chronic pancreatitis diagnoses in the Danish National Patient Registry: A validation study. Scand J Public Health. 2020 Feb;48 (1):14-19. doi: 10.1177/1403494818773535. Epub 2018 May 3.
8. Lidegaard O, Vestergaard CH, Hammerum MS. [Quality monitoring based on data from the Danish National Patient Registry]. Ugeskrift for laeger. 2009;171(6):412-415.
9. Jacobsen TN, Nohr EA, Frydenberg M. Selection by socioeconomic factors into the Danish National Birth Cohort. European journal of epidemiology. 2010;25(5):349-355.
10. Kesmodel US. Information bias in epidemiological studies with a special focus on obstetrics

Saji et al. intended a prospective study to know the effects of warfarin therapy and direct oral anticoagulants (DOACs) on cerebrovascular diseases and cognitive impairment (CI) in patients with non-valvular atrial fibrillation (NVAF) over an estimated duration of 36 months (1). I want to present some information.

First, Mongkhon et al. investigated the risk of dementia/CI among newly diagnosed atrial fibrillation (AF) patients with and without oral anticoagulation (OAC) treatment over a mean follow-up of 5.9 years (2). Hazard ratios (HRs) (95% confidence intervals [CIs) of OAC treatment and antiplatelets for dementia/CI were 0.90 (0.85-0.95) and 0.84 (0.79-0.90), respectively. In contrast, HR (95% CI) of DOACs treatment against warfarin for dementia/CI was 0.89 (0.70-1.14). Furthermore, HR (95% CI) of dual therapy (OAC plus antiplatelets) for dementia/CI was 1.17 (1.05-1.31). Saji et al. set two clinical outcomes, and study period was shorter. In addition, safety evaluation on bleeding might also be required.

Second, Diener et al. recommended randomized trials to evaluate whether OAC, including warfarin and DOACs, reduces dementia/CI in AF patients (3). Dementia/CI, including vascular dementia and Alzheimer's disease, might be related to ischemic stroke, cerebral micro-infarcts, cerebral hemorrhage, and reduced cerebral blood flow. Taken together, risk assessment of medications for dementia/CI in AF patients would be closely associated with types of cerebral disease and its progression. A meta-analysis of randomized trials would finally present reliable information.

References
1. Saji N, Sakurai T, Ito K, et al. Protective effects of oral anticoagulants on cerebrovascular diseases and cognitive impairment in patients with atrial fibrillation: protocol for a multicentre, prospective, observational, longitudinal cohort study (Strawberry study). BMJ Open. 2018;8(11):e021759. doi: 10.1136/bmjopen-2018-021759
2. Mongkhon P, Fanning L, Lau WCY, et al. Oral anticoagulant and reduced risk of dementia in patients with atrial fibrillation: A population-based cohort study. Heart Rhythm. 2020 Jan 10. doi: 10.1016/j.hrthm.2020.01.007
3. Diener HC, Hart RG, Koudstaal PJ, et al. Atrial fibrillation and cognitive function: JACC review topic of the week. J Am Coll Cardiol. 2019;73(5):612-619. doi: 10.1016/j.jacc.2018.10.077

Petrie et al. examined the relationship between average weekly working hours and junior doctors' mental health (1). Common mental disorder (CMD) was assessed using a cut-off of 4 as a threshold on General Health Questionnaire 28-item scale score. Suicidal ideation (SI) was assessed with a single item. Adjusted OR (95% confidence interval [CI]) of working over 55 h/week against working 40-44 h/week for CMD and for SI were 2.05 (1.62 to 2.59) and 2.00 (1.42 to 2.81), respectively. I have some concerns about their study.

First, Soares and Chan described stress levels and the psychological wellbeing of current junior medical officers (JMOs) (2). They used the Short Form-36 and Perceived Stress Scale-14 (PSS14), and JMOs were more likely to have a high PSS-14 score or to have a low mental health score if they reported higher career anxiety. Working hours are closely associated to sleeping time and other lifestyle behaviors, and these factors should be considered for the risk assessment of stress and psychological wellbeing of JMOs.

Second, Rosta and Aasland conducted a survey to experienced doctors to examine the perception of working hours for postgraduate training doctors (3). The question is “How do you think that self-reported total weekly working 45 hours is too short, sufficient, or too long to meet the quality requirements of obligatory postgraduate training for junior doctors?” Although the majority perceived as sufficient for obligatory postgraduate specialist training, doctors of higher age, senior doctors and doctors working in surgical specialties perceived an average of 45 h per week for juniors too short. Although there were specialty differences in perception of sufficient training time, health support for mental disorder and suicidal ideation is indispensable by controlling working hours.

Additionally, Rodriguez-Jareño et al. surveyed the scientific evidence on the health effects of exposure to working hours for doctors (4). Although there is an association of percutaneous injuries and road traffic accidents with extended long working hours, the evidence was insufficient for mood disorders and general health. In addition, dose-response relationship and health risk threshold of extended working hours for physicians have not been determined. Prospective studies are needed for the management of working hours for medical specialists, including junior doctors.

References
1. Petrie K, Crawford J, LaMontagne AD, et al. Working hours, common mental disorder and suicidal ideation among junior doctors in Australia: a cross-sectional survey. BMJ Open. 2020;10(1):e033525. doi: 10.1136/bmjopen-2019-033525
2. Soares DS, Chan L. Stress and wellbeing of junior doctors in Australia: a comparison with American doctors and population norms. BMC Med Educ. 2016;16:183. doi: 10.1186/s12909-016-0693-2
3. Rosta J, Aasland OG. Weekly working hours for Norwegian hospital doctors since 1994 with special attention to postgraduate training, work-home balance and the European working time directive: a panel study. BMJ Open. 2014;4(10):e005704. doi: 10.1136/bmjopen-2014-005704
4. Rodriguez-Jareño MC, Demou E, Vargas-Prada S, et al. European Working Time Directive and doctors' health: a systematic review of the available epidemiological evidence. BMJ Open. 201;4(7):e004916. doi: 10.1136/bmjopen-2014-004916

Thanks to midwife Mette Juhl (MJ) and statistician Per K Andersen (PKA) for their interest in the still hot topic about the risk of stillbirth with increasing gestational age (GA).

MJ and PKA suggest that our calculated cumulative rates of stillbirths by increasing GA are not a good approximation of the cumulative proportion of deaths by increasing GA. The argument for this statement is that there are “competing risks” because “the vast majority of ongoing pregnancies will end in a live birth and only very few in a fetal death”. MJ and PKA prefer to use the cumulative proportions per 1000 deliveries given in Table 1 in our paper.

It is correct that the figures in Fig. 2 are cumulated rates, not per 1000 deliveries but per 1000 ongoing pregnancies. So, these curves demonstrate the risk of fetal death as a function of number of women at risk. We think this way of presenting the increasing risk of fetal death by increasing GA should be just the cumulated incidence rates as a function of the risk time (GA). These curves are thus cumulated rates and should probably have been called that in the publication (instead of cumulated proportion). We do not agree that these cumulated rates are misleading as a documentation of the dramatically increasing risk of fetal death when pregnancy goes weeks beyond term.

The cumulative proportion of deaths per 1000 born is not suited to illustrate the increasing cumulative rate of stillbirth with increasing GA. According to such a calculation the highest proportion of deaths per 1000 born is in week 37. That is because delivery is induced in case of fetal death in week 37, and few women deliver in week 37. Thereby it would seem as if the risk of fetal death is highest in week 37. Therefore, the denominator in all calculations of the risk of death at different gestational ages should be according to the number of women at risk at a certain gestational age, which is the number of pregnant women at that time in pregnancy.

The cumulative proportion, on the other hand, are good to illustrate the decrease in stillbirths by calendar time, and thus the impressing decrease in stillborn per 1000 born by time, actually a decrease from 1.71 per 1000 born from 40 weeks during the period 2000-2008 to 0.99 per 1000 born from 40 weeks during 2009-12, a decrease of 42% and even more so from 41 weeks: A decrease from 1,93 to 0,73 per 1000 born, a decrease of 62%. So far, the more proactive induction practice from around 2010 is by far the most reliable explanation of this fall – not the only contributing factor, but the most important factor.

When women decide at which time they want to be induced having passed 40 weeks, they should know how much the cumulated rates increase by GA. For that purpose, Fig. 2 is illustrative. We have recently updated this curve to the rates prevailing today. They will be published separately.