Of the 1200 records selected, 1184 were available for review. The rates of hypertension as determined by the medical record review, birth data, hospital data, and linked data are presented in . The birth and hospital data underestimated the rates of hypertension. The use of linked data (hypertension in either birth or hospital data) gave rates closer to those in the validation data. Only 12 (0.14%) women in the validation study had chronic hypertension with superimposed preeclampsia.
Rates of hypertension in the validation study and population health data sets (PHDS).
In the validation data, 59 (0.9%) records fulfilled the clinical criteria for severe preeclampsia, and all had a recorded diagnosis of preeclampsia (86%) or pregnancy hypertension (14%). Ten (0.4%) records fulfilled the clinical criteria for chronic hypertension, but no diagnosis was recorded in the medical record and similarly for 7 (0.9%) women with gestational hypertension. Of these, 44% and 9% respectively were reported in one or other PHDS.
Of the medical records reviewed, 642 (56%) were considered less than complete, including 22% without antenatal clinic records and 20% private patients. The rate of chronic hypertension was significantly higher (p = 0.03) in those with complete documentation (2.1%) compared with those with incomplete or no antenatal records (0.7%). However, there was no statistically significant difference in the rate of pregnancy hypertension among women with complete versus incomplete records (9.1 % versus 7.7%, p = 0.4).
The hypertension reporting characteristics of the PHDS compared with the validation data are shown in . For the specific types of hypertension, sensitivities ranged from 23% to 85% and specificities from 96% to 100%. Broad categorization of hypertension into any pregnancy-related hypertension or any hypertension, increased the sensitivity with little impact on the specificity, as did identifying cases for either PHDS. Kappa statistics were also highest for the grouped hypertension categories and for hypertension identified on either dataset. Sixteen of the 25 NPVs (data not shown) were over 98.0%, ranging from 96.9% (95%CI, 95.4% to 97.6%) for gestational hypertension to 100% for preeclampsia identified from either dataset. There was a tendency for the hospital data to be more accurate than the birth data, with generally higher PPVs, NPVs and kappa statistics for all the hypertensive disorders. However, with the exception of preeclampsia reporting, the differences were not statistically significant.
Characteristics of PHDS reporting compar ed with the true occurrence of disease as abstracted from the medical record.
also shows the pregnancy hypertension reporting characteristics of the PHDS compared with the validation data stratified by maternal age and parity. Among the study population 18.2% of women were aged ≥35 years, and 39.9% were having their first baby. The reporting sensitivities and kappas were higher for women aged <35 years but the PPVs were lower, and the differences in reporting were not statistically significant. In contrast, the PPV and NPV were significantly higher for primiparous women than for multiparous women when pregnancy hypertension was identified in either dataset.
In the validation data, 2 women (1.5/10,000) had eclampsia and while both were identified in the hospital data, there were also 6 false-positives (2 with preeclampsia and 4 with gestational hypertension but no record of convulsions) giving a PPV of only 23.5%.
All women identified as having severe preeclampsia in the validation data were reported with pregnancy hypertension in the hospital data and 87% were reported in the birth data, higher than the overall sensitivities for pregnancy hypertension (68% and 63% respectively); under-reporting was concentrated among the less severe forms of pregnancy hypertension. Thus the PHDS included a higher proportion of women with severe preeclampsia than the ‘true’ rate in the validation data, although the differences were not statistically significant (). Women with a preterm birth and/or maternal morbidity were also more likely to have their pregnancy hypertension reported in the PHDS (). Restricting analyses to women with pregnancy hypertension reported in both PHDS datasets resulted in an over-representation of severe disease (as indicated by the highest rates of adverse outcomes) and under-representation of milder disease or women who were well managed. Accepting a report of pregnancy-related hypertension from either data set identified the greatest number of women with pregnancy-related hypertension and had rates of adverse outcomes closest to the ‘true’ rates.
Rates of validated adverse outcomes by source of pregnancy hypertension report.
Finally, we examined the 100 (6%) records where the PHDS had discordant coding of hypertension including 3% where the hypertension was reported in only one dataset. The hospital data included 9 (0.7%) with ‘unspecified maternal hypertension’, and the majority of these (98.5%) were identified as gestational hypertension in the validation study. Overall, the hospital data were more reliable; of the discordant records, the hospital data were correct or partially correct for 76% compared with only 22% for the birth data (p < 0.05). Using the four published methods of classifying discordant records (17) gave odds ratios (OR) for pregnancy-related hypertension and maternal morbidity that ranged from 3.1 to 4.7 compared with the ‘true’ estimate of 3.4 (95%CI, 1.2 to 9.5) although the confidence intervals of all estimates overlapped. Restricting the analysis to records with perfect agreement on hypertension status gave the most extreme odds ratio (4.7; 95%CI, 1.2 to 17.8) while including perfect and partial agreement gave an estimate that was closest to the ‘truth’ (3.2; 95%CI, 1.0 to 10.8).