Four states in Mexico were purposively selected for the study, having large rural and indigenous populations and with a patent need for health systems strengthening to achieve specific health-related Millennium Development Goals. The states of Mexico, Guerrero, Oaxaca and Veracruz have a combined population of 27.7 million or 26.9% of the national total, of which 13.6% is indigenous population, a figure that for Oaxaca rises to 22.5%. They are ranked at the bottom of 32 Mexican states in the Human Development Index, save for the state of Mexico, which ranks 19. State Ministry of Health decision makers and experts selected maternal mortality as a common concern for capacity strengthening through the use of evidence. Indeed, the four states account for 28% of the 1218 cases reported for the country in 2009 and with maternal mortality rates per 100,000 live births for the same year ranging between 106.0 in Guerrero and 45.6 in the state of México, as against 56 nationally. The states of Mexico and Veracruz also account for the largest number of maternal mortality cases of any state.
A community of practice was identified within each state, defined as a group of individuals engaged in roles and relationships to create maternal health system knowledge, define a field of expression and research and identify tools and objects for manipulation . The community of practice in each state was limited for practical purposes to around twenty-five participants, while national decision makers and researchers were also included. In aggregate, the sample consisted of a balanced mix of 11 state and 11 federal decision makers (ministers of health, medical and finance/administration directors, delegates for other federal medical programs); 19 system-wide support staff in charge of financing, human resources, planning, quality of care and teaching, research and training; 19 reproductive health program managers and local area health managers in the poorest municipalities with high incidence of maternal mortality; 22 hospital staff in charge of maternal health, and 12 researchers from academic institutions.
A series of three workshops was held in each state between March and April 2010 combining capacity strengthening for evidence utilization and concept mapping. Federal decision makers and researchers participated individually at a later stage. The first set of workshops in the series was attended by 21 participants per state, on average (86 total, with a minimum of 8 in Oaxaca). Within each workshop, a variable number of small groups of participants with similar organizational profiles brainstormed to the question "Which are the health system problems that block access to interventions and tools of proven effectiveness to promote maternal health, prevent disease during pregnancy and avoid maternal mortality?" Ideas were first produce in writing and then read aloud by a moderator to reduce duplicates through discussion.
The 460 ideas produced by small groups were later integrated into 99 unique problems through content analysis and generalization, bringing them to a number that could allow reliable ranking and sorting in the second workshop . Problems were individually printed in 2 × 3 cards with five-point rating scales to measure importance for maternal health ("not important" to "of vital importance") and feasibility of being solved ("impossible"; "difficult"; "possible but no solution known"; "a solution is being formulated"; "a solution is being implemented"). Stacks of cards were distributed to 79 of the original 86 participants and to 15 federal decision makers and researchers. After explaining the purpose of the exercise, each participant proceeded to rank each problem and to sort them into as many piles as they thought important to classify problems for their strategic analysis. The only rules were to produce more than one pile and to allocate each card to only one pile. Participants were asked to name each pile with a descriptive title and to place piles in a named envelope.
Pile data and sorter's role and gender were input to the software Concept Systems Core@ V.4 . Problems were mapped through multi-dimensional scaling and factor analysis for groups with at least 11 sorters to ensure reliable maps . Specific problems were mapped in diametrically opposed positions when they were never or seldom grouped together in the card piles across sorters; problems were mapped more tightly together when they were often piled together, and were mapped in the middle when they were as often as not sorted together. Factor analysis grouped the problems into as many function regions as desired at different levels of aggregation.
Data for all participants was first used to generate national concept maps. Actor data was then processed separately to generate and rank six specific maps, one for each actor at the national level.
After analyzing between 8 and 15 levels of aggregation for the national map, a ten-function map was selected as the most significant and parsimonious. Specific maps across actors were generated at the same level for comparative purposes. The Concept Mapping software assigned the most encompassing label to each function, drawing from the individual pile labels and choosing the one with best statistical fit. Function labels for state-level maps were in some cases modified by state communities of practice at a third workshop, seeking the fullest agreement they thought possible with as many of the specific problems contained within each function. Labels for the actor-specific maps were likewise reviewed and modified by the project team.
Maps and functions within them were analyzed for degree of classification, importance, feasibility and structural position. Degree of classification is the number of problems contained within each function. The larger the number of problems, the more richly classified is the function. The Concept Mapping software ranked each of the ten health system functions by importance and feasibility using the problem rankings of individual participants. The software did this by dividing the average grade differences across functions by five, and assigning a value of one to the lowest rank and a value of five to the highest. Spearman correlations were then run by the software across chosen actors for all ten functions.
Structural relationships between functions were compared within and across maps. To this end, the most commonly identified function across all maps was identified and used as reference to align all maps with this function at the top-centre position. Quadrants were then identified and the functions within each quadrant were grouped and placed as bars in a column, with functions at the centre of maps placed at the bottom. By placing columns besides each other, concept maps could then be compared.
The research protocol was reviewed by the Ethics Committee of the National Institute of Public Health. No informed consent was required. All data was maintained confidential and names of subjects were deleted after their profiles were entered in the database.