Beyond signal functions in global obstetric care: Using a clinical cascade to measure emergency obstetric readiness

Background Globally, the rate of reduction in delivery-associated maternal and perinatal mortality has been slow compared to improvements in post-delivery mortality in children under five. Improving clinical readiness for basic obstetric emergencies is crucial for reducing facility-based maternal deaths. Emergency readiness is commonly assessed using tracers derived from the maternal signal functions model. Objective-method We compare emergency readiness using the signal functions model and a novel clinical cascade. The cascades model readiness as the proportion of facilities with resources to identify the emergency (stage 1), treat it (stage 2) and monitor-modify therapy (stage 3). Data were collected from 44 Kenyan clinics as part of an implementation trial. Findings Although most facilities (77.0%) stock maternal signal function tracer drugs, far fewer have resources to practically identify and treat emergencies. In hypertensive emergencies for example, 38.6% of facilities have resources to identify the emergency (Stage 1 readiness, including sphygmomanometer, stethoscope, urine collection device, protein test). 6.8% have the resources to treat the emergency (Stage 2, consumables (IV Kit, fluids), durable goods (IV pole) and drugs (magnesium sulfate and hydralazine). No facilities could monitor or modify therapy (Stage 3). Across five maternal emergencies, the signal functions overestimate readiness by 54.5%. A consistent, step-wise pattern of readiness loss across signal functions and care stage emerged and was profoundly consistent at 33.0%. Significance Comparing estimates from the maternal signal functions and cascades illustrates four themes. First, signal functions overestimate practical readiness by 55%. Second, the cascade’s intuitive indicators can support cross-sector health system or program planners to more precisely measure and improve emergency care. Third, adding few variables to existing readiness inventories permits step-wise modeling of readiness loss and can inform more precise interventions. Fourth, the novel aggregate readiness loss indicator provides an innovative and intuitive approach for modeling health system emergency readiness. Additional testing in diverse contexts is warranted. Kakamega County was selected by the Government of Kenya for a parent implementation trial testing the impact of a package of community- and clinical quality interventions on the uptake and quality of facility-based care. Kakamega was selected by the MoH based on government data indicating the county’s MMR of 800 was nearly double the country’s ratio of 413 [44, 47]. Forty four facilities and their catchment areas were selected for intervention using four criteria: primary care clinics with KEPH Level 2–3 designation (Kenya Essential Package of Health [47]), providing basic emergency obstetric services as defined by the MoH (BEmOC) [48], conducting 10 or more deliveries in the previous calendar year (2011) and being located in one of five sub-counties within Kakamega County (for the purpose of analysis, two facilities that were formerly part of Kakamega Central prior to post-constitutional rezoning were retained in the study based on the intent-to-treat principle. Thus, the results report on five sub-counties since the newly designated Navakholo county was retained with Kakamega Central for the parent trial. The cross-sectional analysis of facility readiness is nested within a non-equivalent group design pre-post implementation trial evaluating a facility- and community-intervention package in Kakamega County, Kenya [49]. In the parent trial, 756 facility-specific variables were collected at 44 primary care facilities; this nested study used 80 obstetric-specific variables collected from facilities prior to the start of the intervention. Obstetric emergency readiness at the facility-level has been defined by the proportion of specified clinical items that are present at a facility on the day a facility inventory is conducted [33]. Although there is no universal consensus on the number of tracers that should be used to measure emergency readiness as defined by the signal functions model [30–33, 50], WHO’s Service Readiness Index (SRI) defines basic emergency obstetric readiness using 7 tracers (composed of 9 discrete items). These tracers are measured using observation and/or interview during facility visits [33]. This signal function-based approach to estimating emergency readiness uses 3 parenteral drugs (uterotonic, antibiotics, anticonvulsant), 3 intravenous items (including IV solution and a 2-part IV infusion kit), 1 durable good (manual vacuum apparatus) and 2 multipurpose items (gloves and light source). The WHO-SRI standardized tool was used to create a signal function-based estimate of emergency readiness at the 44 primary care clinics. Next, we measured readiness using a novel emergency obstetric clinical cascade model derived from Potter’s hierarchy of needs framework [40] and the HIV care cascade model [42]. The resulting obstetric clinical cascade quantifies resources required to sequentially identify, treat and manage basic obstetric emergencies as they present clinically at primary care facilities. Consequently, emergency obstetric readiness is reported as the percentage of facilities with all of the related clinical tools for managing obstetric emergencies (as defined by the two models). The higher the percentage is, the higher the facilities’ readiness is to manage basic obstetric emergencies. Although facility-level estimates of readiness could be calculated, this study reports the percent readiness aggregated across all 44 clinics. The current signal function readiness estimates are reported for all basic maternal emergency obstetric signal functions as a single indicator—the proportion of facilities with tracer items for all manual procedures and all medical treatments. Standard signal function estimates and the WHO’s obstetric service readiness index (SRI) do not measure readiness for each clinical disorder. However, if one estimated emergency-specific readiness using the signal function tracer items alone, many resources required to practically deliver care would be absent. For example, signal function estimates for eclampsia would be defined as the proportion of facilities with IV solution/infusion set, hydralazine and magnesium sulfate [33]. Using these three items to model eclampsia emergency readiness alone does not account for the resources required to first identify if the emergency is present (i.e., sphygmomanometer, stethoscope, urine collection device and urine protein test). Also, it does not explicitly model all necessary drugs or ancillary resources required to practically deliver the first-line treatment. Although consumable supplies are required to deliver treatment drugs, consumable resources are often omitted from the signal functions for most emergencies (for example, the required refrigeration for oxytocin is not modeled in the oxytocic function and IV tubing, IV catheter and IV solution are discrete, interdependent items that are only measured as one item in signal function estimates. Consequently, reporting capacity when one or more of the three interdependent items are missing is not precise or accurate). Further, although durable goods are essential for conducting procedures or delivering drugs, they are often excluded from signal function estimates (i.e., IV poles, syringes or needles for delivering drugs). For some emergencies, the specific drug required for the emergency is not modeled by the signal functions (For example, in the oxytocic signal function, the cause of post-partum hemorrhage (PPH) and the varied drugs for treating hemorrhage based on its underlying cause are not modeled). In contrast, the proposed cascade model is a clinically-oriented approach to measuring readiness. It is based on a practical, step-wise cascading relationship between resources [40, 42]. The resources for identifying the emergency (Stage 1) are required first before accurate treatments can be administered to patients (Stage 2). Further, the cascade explicitly models the consumable supplies and durable goods required to practically deliver treatment drugs in clinical practice (for example, in eclampsia, the cascades model the interrelationship between all clinical resources required to first identify the disorder and then deliver the treatment drug). Thus, emergency readiness in the cascade model is the proportion of facilities with the treatment drug that can first identify the disorder (stage 1, Identify) and then have the durable and consumable resources to administer the treatment drug (stage 2, Treat). Since the signal functions do not measure care quality, the third cascade stage for monitoring and modifying therapy as clinically indicated (Stage 3) is not used to compare the signal functions and cascade models (Fig 1). Each emergency cascade’s title is based on the underlying clinical disorder and paired with existing signal functions as follows: Manage Sepsis-Infection (the parallel signal function is parenteral antibiotics), Hemorrhage (oxytocics), Hypertensive Emergency (anticonvulsant), Retained Placenta (manual removal of retained placenta), Incomplete Abortion (removal of retained products of conception) [22]. A facility’s ability to monitor or modify the primary treatment based on a patient’s clinical response (Stage 3) is a proposed indicator for measuring clinical quality but not for evaluating signal function performance (Fig 2). Tracers were precisely defined to minimize ambiguity in signal function estimates and to facilitate comparison between models. When tracers were not explicitly defined by the signal functions, the WHO first-line recommendations for obstetric care were used [23, 51, 52] to make a general tracer from the signal function model (i.e., parenteral antibiotics) more specific (i.e., parenteral ampicillin and/or penicillin alternative). Four other ambiguous tracers included light source, IV supplies, drugs and emergency protocols. Light was defined as functional electric lights or functional flashlights. The IV kits tracer includes three discrete resources: drug-compatible fluids, tubing and a venous access device/cannula. We modeled this using the cannula and fluids since data on tubing were absent. Fluid was modeled as Lactated Ringer’s/Hartman solution or normal saline since both fluids are compatible with emergency drugs available at the study facilities. For the cascades, the first-line clinically-indicated drug was used to model readiness when the drug was not defined by the signal function model. The oxytocics signal function does not specify the precise utertonic drug required for various hemorrhage emergencies. When managing obstetric hemorrhage, the preferred first-line drug is oxytocin [23, 51]. However, in its absence, misoprostol or ergometrine could be substituted (with blood pressure monitoring). In contrast, when managing incomplete abortion, the preferred uterotonic is a non-oxytocin agent such as ergometrine or misoprostol. Therefore, the presence of any first- or second-line uterotonic specific to the emergency was used to model readiness in the cascade model since the cascades are based on specific clinical emergencies [23, 51]. Further, the parentral antibiotic signal function does not define the tracer antibiotic drugs required. We used the WHO’s 3-step sequence of obstetric antibiotic therapy escalation based on the type of suspected infection to define readiness: Step 1—ampicillin, 2—gentamicin and 3—metronidazole [51]. Since most facilities lack ampicillin, the presence of ampicillin or any of three alternative penicillin drugs (benzathine, procaine or crystalline) was used to model this WHO step 1 antibiotic readiness (the primary study’s clinical inventory did not capture metronidazole availability, so WHO’s step 3 antibiotic readiness was not modeled). Although some emergency protocols are tracers in the signal function model, they were selectively used to model the quality of clinical care but not general emergency readiness for four reasons: 1) individual clinician knowledge and skill vary so having a protocol does not guarantee readiness, 2) by extension, protocol absence does not guarantee a lack of emergency readiness, 3) signal functions do not define the protocol required [33] and 4) several protocols may actually be required to manage the primary emergency’s sequella. For example, when a patient presents with a retained placenta, a subset of patients may develop post-partum hemorrhage (PPH), endometritis and/or sepsis while another subset may resolve with first-line treatment. Identifying some clinical disorders is based primarily on clinician skill. Although clinician skills vary widely [53], a 100% skill level was assumed for all cascades since skill assessment is not include the signal function estimates of facility emergency readiness [22, 33]. Some items required for comprehensively modeling readiness in all six maternal signal functions were absent from the baseline facility inventory. This analysis focuses on three basic medical and two manual functions since data on assisted vaginal delivery supplies were absent [21, 22]. However, the expanded cascades in supplemental tables include resources for all six maternal cascades. In parallel with the WHO- readiness estimates based on the signal functions, this study measured the cross-sectional availability of routine and emergency obstetric resources during facility visits conducted by study staff between February and May 2013. Three trained research assistants used standardized forms to visually identify emergency resource availability and ask clinic mangers about resource availability when items were not initially located using observation. This method of survey data collection matches the WHO-SRI approach used to quantify signal function estimates of emergency readiness [33]. 80 variables from the inventory describe facility demographics, staff, consumable medical supplies, durable goods and obstetric drugs. Mean estimates of maternal signal function readiness are derived from 396 observations (9 tracer items from 44 facilities). Cascade estimates of readiness utilized 1,364 observations (31 variables from 44 facilities). One author (JD) trained all staff on this the facility inventory instrument. The author also provided periodic in-person and remote instrument coaching and data quality assurance in-services. A trained clerk entered these data into the RedCap’s online database (Institute for Translational Health Sciences, 2007–2015). Accuracy of these data were confirmed using a standard double-entry technique where two assistants entered data from one quarter of the paper forms. Data clerks resolved any discrepancies between the two RedCap entries by reviewing the original paper forms. Thus, any discrepant REDCap entries were reconciled with the original paper records. The resulting validated database was used for analysis. RedCap data were exported to STATA for analysis (version 11.2, College Station, Texas, 1985–2009). We described obstetric variables with standard descriptive statistics; point estimates for the availability of each resource are reported as percentages. Since the variables in this dataset had fewer than 100 observations, skewed distributions or did not follow symmetric Gaussian distribution, non-parametric descriptive and inferential statistics with two-sided tests significance were used for all analyses. Drop-offs in readiness between each stage were quantified with percentages. Central tendency was typically reported as the median. Means were used primarily for estimates of overall emergency readiness estimates in both models for two reasons: 1) the SRI methodology uses means and 2) since this measure is based on few observations the median would not capture the range of observations effectively or accurately. Variability was primarily summarized using absolute ranges since facilities varied widely in the obstetric resources available. Standard deviation (SD) and interquartile ranges (IQR) were selectively used as measures of variance when variation in the central tendency and range was of interest (for example using both metrics for the number of monthly deliveries illustrates wide variability in delivery volume by study site). Since global variability in urban-rural obstetric care is well-documented [3, 39, 54–56], we statistically quantified periurban/rural differences in the facility characteristics based on a facility’s rural status using Kenya MoH definitions. To test differences between proportions, we used Pearson’s chi-square test of independence or Fischer’s exact test (for cell counts less than five). When comparing a variable’s distribution across unpaired categories, we used Wilcoxon-ranked sum test (for two categories) or Kruskal-wallis’ h-test (more than two categories). We used the unmatched median test to compare medians across two unpaired categories. The ‘signal function overestimate’ indicator is calculated by subtracting the novel cascade estimate of readiness from the standard signal function estimate of readiness (signal function estimate [–] clinical cascade estimate [=] readiness overestimate by signal function). The activities and analysis of this nested study were all contained in the parent study approved by the University of Washington Institutional Review Board (43069) and the University of Nairobi Ethical Review Committee (P57/05/2012). The trial is registered in the PanAfrican Clinical Trials Registry (PACTR0121200045732, available from: http://www.pactr.org). Since the intervention targeted clinics and not individual clinical providers, prior to the clinic-level intervention, individual clinicians were verbally informed of the study and provided the opportunity to opt-out of the clinical training or assessments; no one opted out. Further, the MoH provided authorization to collect these data at the MoH facilities as part of the implementation trial. Consequently, the facility inventory data did not require individual informed consent.