We conducted a descriptive cross-sectional study using a mixed methods approach, whereby we triangulated qualitative and quantitative research methodologies [6]. We obtained ethical approval from the Mount Kenya University Ethical Review Committee (approval number 1309) and research accreditation from the National Commission for Science, Technology and Innovation (license number NACOSTI/P/22/19461).

The target population was 2521 people: 1500 community health workers, 120 nurses, 570 maternity patients, 36 clinical officers in charge of health facilities, and 288 health facility management committee members, along with 7 county and sub-county reproductive health coordinators from 6 sub-counties in Mombasa County. The community health workers are volunteers, supervised by community health extension workers, who worked in 179 community health units of around 100 families or 5000 community members, which are the first tier in a four tier system of health care delivery in Kenya. All the other participants except the 7 reproductive health coordinators, whom we included as key informants, worked in 36 facilities, all levels 2 and 3 public health facilities. These facilities are in the second tier of the health care system, are under the control of the county government, and are made up of primary care health facilities with dispensaries and health centers staffed by nurses and clinical officers. The third and fourth tiers of the health care system are made up of county referral hospitals and national referral hospitals respectively.

We calculated a sample size of 345 respondents from the remaining population of 2514 using Yamane’s formula [7]:

$$} = } \div \, ( + \, \left( } \times \, 0.0^} } \right),$$

(1)

where N is the population size; n is the sample size; and 0.05 is the precision rate. We then adjusted upwards to 383 respondents to accommodate a 10% probable withdrawal or non-response rate:

$$}_} = } \div \, \left( - \, 0.} \right),$$

(2)

where n1 is the adjusted sample size; n is the calculated sample size; and 0.1 is the estimated non-response rate. We then included all the 36 clinical officers in charge of the levels 2 and 3 facilities, and used proportionate stratified random sampling to sample 17 nurses, 210 community health workers, 80 maternity patients, and 40 health facility management committee members.

We then used systematic random sampling to select respondents from the respective strata to provide the members in each stratum equal opportunity to participate in the study [8]:

$$} = \, \left( }_} \div }} \right) \, \times },$$

(3)

where N is the population size; and n1 is the adjusted sample size. The final sample consisted of 390 participants, including 7 key informants.

We used a structured questionnaire, a modified Quality of Decision-making Orientation Scheme (QoDoS) [9,10,11,12,13], and a Key Informant Interview (KII) guide to collect data. We attached an informed consent form to the data collection tools and respondents voluntarily completed them prior to data collection. We used a drop-off and pick-up method to administer the study questionnaire and the modified Quality of Decision-making Orientation Scheme. We used the questionnaire to assess the independent variables including; the frequency of utilization of participatory needs assessment, baseline assessments, and analysis of objectives at the initiation phase in maternal and newborn health programs (using questions 13 to 15, as listed in the Supplementary Material); the frequency of utilization of participatory feasibility analysis, Strengths, Weaknesses, Threats and Opportunities (SWOT) Analysis, and risk assessment at the design and planning phase of the programs (using questions 16 to 18, as listed in the Supplementary Material); and the frequency of utilization of participatory performance reviews, desk reviews, and supportive supervision at the implementation phase of the programs (using questions 19 to 21, as listed in the Supplementary Material).

We used the modified scheme to assess the frequency of quality decision-making practices (QDMPs) at the individual level and organizational levels of the health facilities with four indicators of quality decision-making. These include decision-making in terms of approach, culture, competence, and style [13]. These four indicators are based on 10 quality decision-making practices that this study has adopted as a catalog of ideal elements of quality decision-making at the health facilities. These elements include having a systematic, structured approach to aid decision-making (consistent, predictable and timely); assigning clear roles and responsibilities; assigning values and relative importance to decision criteria; evaluating both internal and external influences/biases; examining alternative solutions; considering uncertainty; re-evaluating as new information becomes available; performing impact analysis of the decision; ensuring transparency and provide a record trail; and effectively communicating the basis of the decision [9]. At the organizational level, we assessed the practices using two indicators of the modified scheme: decision-making approach and culture (using questions 22 to 41, as listed in the Supplementary Material). At the individual level, we evaluated the practices using the other two indicators: decision-making competence and style (using questions 42 to 65, as listed in the Supplementary Material).

The study outcomes are self-reported frequencies, for which high frequencies are perceived indications that the health facilities used participatory approaches and engaged in quality decision-making practices to a great extent. We measured the four self-reported indicators of quality decision-making using sets of 5-point Likert scales (1 to 1.8—Not at all; 1.81 to 2.6—Sometimes; 2.61 to 3.4—Frequently; 3.41 to 4.2—Often; 4.21 to 5—Always). We aggregated, using arithmetic mean, the self-reported frequencies of decision-making approach and culture, and decision making competence and style, to measure the perceived quality decision-making at the organization level and individual level respectively. We further aggregated, using arithmetic mean, the scores for quality decision making at the organization level and individual level to obtain self-reported scores of perceived quality-decision at the health facilities’ maternal and newborn programs.

We measured utilization of the nine participatory approaches, three at each program phase including initiation, design and planning, and implementation, using a 5-point Likert scale (1 to 1.8—Not at all; 1.81 to 2.6—Sometimes; 2.61 to 3.4—Frequently; 3.41 to 4.2—Often; 4.21 to 5—Always). We then aggregated, using arithmetic mean, the self-reported frequencies of the participatory approaches, each aggregated set comprising three approaches in a program phase, to measure the perceived utilization of participatory monitoring and evaluation approaches at the initiation, design and planning, and implementation phases.

We used the final aggregated self-reported scores to assess the relationship between the independent variables and the perceived quality decision making in the programs. To improve the validity of the model, we merged the five response categories in the 5-point Likert scale to obtain two response categories, ‘rarely’ and ‘often’. To determine the category intervals, we subtracted the lowest point in the 5-point Likert scale from the highest point and divided the difference by the required number of categories [(5–1)/2]. Therefore, we recoded the aggregated self-reported scores ranging between 1 and 3 to a response category named ‘rarely’ and recoded scores ranging between 3.1 and 5 to a response category named ‘often’. The category ‘rarely’ signified that the extent of use or practice was minimal while the category ‘often’ signified that the extent of use or practice was great.

We performed descriptive analyses (using arithmetic mean, standard deviation and coefficient of variation) to summarize data on participants' demographics and specific variables and binary regression analysis (at a significance level of 0.05) to detect relationships between the selected frequency in individual questions and variables and the dependent variable. We conducted a Phi coefficient test, at a significance level of 0.05, to determine the strength of association between the frequency of quality decision making practices (dependent variable) and the independent variables, such as the frequency of utilization of participatory monitoring and evaluation approaches at the initiation phase, frequency of utilization of participatory monitoring and evaluation approaches at the design and planning phase, and frequency of utilization of participatory monitoring and evaluation approaches at the implementation phase. The Phi coefficient ranges from − 1 to + 1 with a negative coefficient signifying negative relationship, zero signifying no relationship, and a positive coefficient signifying positive relationship between the dependent variable and an independent variable. A Phi coefficient greater or less than zero with a p-value (significance level) less than 0.05 was deemed to indicate significant association between an independent variable and the outcome variable.

We used a binary logistic regression analysis to determine the perceived degree of influence of participatory monitoring and evaluation approaches at the initiation, design and planning, and implementation phases on self-reported frequency of decision-making (at 0.05 significance level). We set the last response category as the reference group. The Hosmer–Lemeshow test yielded a significance value greater than 0.05 indicating that the model adequately fit the data: that is, there was no difference between the observed and predicted models. The results also indicated that the model correctly classified 65.3% of cases. The Nagelkerke R2 test indicated that the model (utilization at the initiation, design and planning, and implementation phases) explained 20.9% of the variance in quality of decision-making. An odds ratio of 1.0 indicated that an independent variable was not associated with quality decision-making (dependent variable). An odds ratio of greater than 1.0 indicated that the independent variable was a catalyst for quality decision-making practices. An odds ratio of less than 1.0 indicated that the independent variable was an inhibitor of quality decision-making practices. An odds ratio greater or less than 1.0 with a p-value (significance level) less than 0.05 was deemed to indicate significant influence of the independent variable on the outcome variable.

We conducted the key informant interviews on the same day and recorded all sessions using a digital voice recorder, then transcribed them verbatim. Each interview session lasted about 60 min. To assess qualitative data from the interviews we used content analysis. The IBM statistical package for the social sciences (for Windows), version 25, was used for quantitative analysis, while content analysis was conducted manually.